Keywords: Bio-electric sensors - Sensing methods, Bio-electric sensors - Sensor systems
Abstract: Cancer recurrence is an important issue in bladder tumor resections, because tissue cannot generously be removed from the thin bladder wall without impacting bladder functionality. Electrical impedance measurements during an operation aim to support the surgeon in making the decision which tissue areas to preserve and which ones to remove, because structural and physiological changes in tissue due to cancerous cell aggregations can be detected by their altered electrical characteristics This work investigates the detection limits of tetrapolar impedance sensors when the impedance of heterogeneous tissue is measured. To do this, a finite element analysis is carried out where the sensors are placed on a dielectric medium with inclusions of different conductivities and at different locations relative to the sensor. It is shown that a sensor with four electrodes in a square performs poorly in comparison to a sensor where the electrodes are symmetrically shaped as rings around one center electrode. This is mainly due to its enlarged regions of negative sensitivity. Based on the results, a third, optimized sensor geometry is proposed that shows superior performance to the other sensors in terms of geometry factor, sensitivities, and tumor detection. In simulation, it can reliably detect tumors with only half the radius of the sensor surface. Smaller tumor fractions, however, cannot be detected by either sensor.

Keywords: Bio-electric sensors - Sensor systems, Wearable sensor systems - User centered design and applications, Optical and photonic sensors and systems
Abstract: The application of transparent conductive films to flexible biomedical optoelectronics is limited by stringent requirements on the candidate materials' electromechanical and optical properties as well as their biological performance. Thin films of graphene and poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) are sought as mechanically flexible alternatives to traditional indium tin oxide (ITO). However, they require more understanding of their suitability for biomedical optoelectronic devices in terms of transmission behavior and electromechanical stability. This study shows that the relative increase in sheet resistance under cyclic loading for ITO, graphene, and PEDOT:PSS was 3546±3908%, 12±2.7%, and 62±68%, respectively. Moreover, graphene and PEDOT:PSS showed a transmission uniformity of 9.3% and 36.3% (380-2000 nm), respectively, compared with ITO film (61%). Understanding the optical, electrical, and mechanical limits of the transparent conductive films facilitates the optimization of flexible optoelectronic designs to fit multiple biomedical research and clinical applications.

Keywords: Wearable sensor systems - User centered design and applications, Health monitoring applications, Integrated sensor systems
Abstract: Drowsiness monitoring can reduce workplace and driving accidents. To enable a discreet device for drowsiness monitoring and detection, this work presents a drowsiness user-study with an in-ear EEG system, which uses two user-generic, dry electrode earpieces and a wireless interface for streaming data. Twenty-one drowsiness trials were recorded across five human users and drowsiness detection was implemented with three classifier models: logistic regression, support vector machine (SVM), and random forest. To estimate drowsiness detection performance across usage scenarios, these classifiers were validated with user-specific, leave-one-trial-out, and leave-one-user-out training. To our knowledge, this is the first wireless, multi-channel, dry electrode in-ear EEG to be used for drowsiness monitoring. With user-specific training, a SVM achieved a detection accuracy of 95.9%. When evaluating a never-before-seen user, a similar SVM achieved a 94.5% accuracy, comparable to the best performing state-of-the-art wet electrode in-ear and scalp EEG systems.

Keywords: Wearable sensor systems - User centered design and applications, Physiological monitoring - Modeling and analysis, Health monitoring applications
Abstract: Swallowing recognition is the leading step in the evaluation of dysphagia which seriously affects people’s life. Current medical swallowing monitoring methods require an in-hospital environment and overly rely on professional knowledge of the medical staff. In this study, we developed a wearable swallowing recognition system that consists of an on-neck wearable swallowing sensing device and a data processing module on a host computer. The wearable device collects inertial signals including acceleration and angular velocity, as well as dual photoplethysmography (PPG) signals based on infrared and green light from the neck. A novel processing framework for dual PPG signals is proposed to extract and enhance the laryngeal motion component introduced by swallowing activities in the data processing module. The laryngeal motion component of dual PPG signals together with the preprocessed inertial signals are further used for feature extraction to proceed swallowing recognition based on random forest classifier. We collected data from 32 healthy subjects in the center and side positions on the neck using our system to analyze their swallowing activities. As a result, we achieved a high average area under curve (AUC) of the swallowing recognition by 86.6%. We also find the sensing position has a significant impact on gender-specific swallowing recognition performance, as the center position was better for females (92.9%), while the side position was better for males (87.6%). The results indicate that the proposed system could achieve high integrity and good performance, which is helpful for the future swallowing research.

Keywords: Optical and photonic sensors and systems
Abstract: Optical tweezer is a non-contact tool to trap and manipulate microparticles such as biological cells using coherent light beams. In this study, we utilized a dual-beam optical tweezer, created using two counterpropagating and slightly divergent laser beams to trap and deform biological cells. Human embryonic kidney 293 (HEK-293) and breast cancer (SKBR3) cells were used to characterize their membrane elasticity by optically stretching in the dual-beam optical tweezer. It was observed that the extent of deformation in both cell types increases with increasing optical trapping power. The SKBR3 cells exhibited greater percentage deformation than that of HEK-293 cells for a given trapping power. Our results demonstrate that the dual-beam optical tweezer provides measures of cell elasticity that can distinguish between various cell types. The non-contact optical cell stretching can be effectively utilized in disease diagnosis such as cancer based on the cell elasticity measures.

Keywords: Coupling and synchronization - Nonlinear coupling, Physiological systems modeling - Signal processing in simulation
Abstract: Cross frequency coupling (CFC) between electrophysiological signals in the brain has been observed and it's abnormalities have been observed in conditions such as Parkinson's disease and epilepsy. More recently, CFC has been observed in stomach-brain electrophysiologic studies and thus becomes an enticing possible target for diseases involving aberrations of the gut-brain axis. However, current methods of detecting coupling do not attempt to capture the underlying statistical relationships that give rise to this coupling. In this paper, we demonstrate a new method of calculating phase amplitude coupling by estimating the mutual information between phase and amplitude, using a flexible parametric modeling approach. Specifically, we develop an exponential generalized linear model (GLM) to model amplitude given phase, using a high dimensional basis of von-Mises function regressors and L1 regularized model selection. Using synthetically generated gut-brain coupled signals, we demonstrate that our method outperforms the existing gold-standard methods for detectable low-levels of phase amplitude coupling through receiver operating characteristic (ROC) curve analysis.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Nonlinear dynamic analysis - Phase locking estimation
Abstract: This study examines power-power cross-frequency coupling (CFC) between different frequency bands of cortical activity in normal-hearing (NH) listeners and its association to the processing temporal envelope (ENV) and temporal fine structure (TFS) of speech. CFC between alpha and theta bands and between gamma and theta bands was investigated when only ENV or TFS or the original speech itself were processed. Comparing the cortical activity in response to ENV and original speech, there was an increase in alpha-theta CFC and in gamma- theta CFC when listening to ENV alone. However, when comparing the response when to listening TFS alone, there was a reduction in gamma-theta CFC compared to the original speech and the alpha-theta CFC was comparable to the equivalent observed with original speech. The increase in CFC may suggest that there is more synchrony across different bands of cortical activity in processing ENV than TFS. These measures can serve as indicators when either ENV or TFS is perceived.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: Graph neural networks (GNN) are an emerging framework in the deep learning community. In most GNN applications, the graph topology of data samples is provided in the dataset. Specifically, the graph shift operator (GSO), which could be adjacency, graph Laplacian, or their normalizations, is known a priori. However we often have no knowledge of the grand-truth graph topology underlying real-world datasets. One example of this is to extract subject-invariant features from physiological electroencephalogram (EEG) to predict a cognitive task. Previous methods use electrode sites to represent a node in the graph and connect them in various ways to hand-engineer a GSO e.g., i) each pair of electrode sites is connected to form a complete graph, ii) a specific number of electrode sites are connected to form a k-nearest neighbor graph, iii) each pair of electrode site is connected only if the Euclidean distance is within a heuristic threshold. In this paper, we overcome this limitation by parameterizing the GSO using a multi-head attention mechanism to explore the functional neural connectivity subject to a cognitive task between different electrode sites, and simultaneously learn the unsupervised graph topology in conjunction with the parameters of graph convolutional kernels.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Causality, Coupling and synchronization - Nonlinear coupling
Abstract: In the last years, the characterization of brain-heart interactions (BHIs) in epilepsy has gained great interest. For some specific seizures there is still a lack of information about the mechanisms occurring during or close to ictal events between the central nervous system (CNS) and the autonomic nervous system (ANS). This is the case for neonatal seizures, one of the most common neurological emergencies in the first days of life. This paper evaluates possible differences in BHIs between newborns with seizures and seizure-free ones. We applied convergent cross mapping approaches to a cohort of 52 newborns from a public dataset. Preliminary results show that newborns with seizures have a lower degree of interaction between the CNS and the ANS than seizure-free ones (Mann-Whitney test: p-value <0.05). These results are of clinical relevance for future BHI-based approaches to better understand the neural mechanisms behind neonatal seizures.

Keywords: Connectivity, Principal component analysis
Abstract: Seizures represent a brain activity state characterised by extended synchronised firing in multiple regions that prevent normal brain functioning. It is important to develop methods to distinguish between normal and abnormal synchronisation in epilepsy, as well as to localise the networks involved in seizures. To this end, we perform a preliminary investigation in the use of principal components analysis (PCA) to assess the change in dynamic electroencephalogram (EEG) connectivity before and after seizure onset. Source estimation was performed for an openly available EEG dataset from 14 patients with epilepsy. By applying PCA onto the EEG data processed into dynamic connectivity (dFC) matrices, we identified a set of connectivity topologies (eigenconnectivities) that explain high levels of variance in the dynamic connectivity. We compare the dimensionality reduction results obtained on source-level vs. scalp-level connectivity. We identified eigenconnectivities with differences in preictal vs. ictal activity and the brain networks associated with these activations. The work illustrates a data-driven approach for identification of topologies of brain networks that change with seizure onset.

Keywords: Connectivity, Directionality, Multivariate methods
Abstract: Isolated effective coherence (iCoh) is a measure of neural causal functional connectivity from EEG signals that was proven to overperforms the Generalized Partial Directed Coherence (gPDC). However, iCoh sensitivity in the identification of reliable functional neural connections with respect to random links was not investigated. This study aims to compare the sensitivity of iCoh and gPDC with a statistical surrogates’ approach. The cerebral motor network topology of a cohort of subjects in sub-acute stage after stroke was investigated. iCoh showed enhanced statistical discriminative power of the relevant connections within the motor network with respect to gPDC. This property influenced the assessment of ipsilesional intra-hemispheric topographic variations occurring in the population after a physical rehabilitation program.

Keywords: Data mining and big data methods - Inter-subject variability and personalized approaches, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Various pattern-recognition or machine learning-based methods have recently been developed to improve the accuracy of the motor imagery (MI)-based brain-computer interface (BCI). However, more research is needed to reduce the training time to apply it to the real-world environment. In this study, we propose a subject-transfer decoding method based on a convolutional neural network (CNN) which is robust even with a small number of training trials. The proposed CNN was pre-trained with other subjects’ MI data and then fine-tuned to the target subject’s training MI data. We evaluated the proposed method using the BCI competition IV data2a, which had the 4-class MIs. Consequently, on the same test dataset, with changing the number of training trials, the proposed method showed better accuracy than the self-training method, which used only the target subject’s data for training, as averaged 86.54±7.78% (288 trials), 85.76±8.00% (240 trials), 84.65±8.11% (192 trials), and 83.29±8.25% (144 trials), respectively, which was 4.94% (288 trials), 6.10% (240 trials), 9.03% (192 trials), and 12.31% (144 trials)-point higher than the self-training method. Consequently, the proposed method was shown to be effective in maintaining classification accuracy even with the reduced training trials.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: In recent years, neuroscientists have been interested to the development of brain-computer interface (BCI) devices. Patients with motor problems may benefit from BCIs as a means of communication and for the restoration of motor function. To implement a BCI, usually electroencephalography (EEG) is used for evaluating the neuronal electric potential activity. In computer vision applications such as image classification and object recognition, deep neural networks (DNN) show significant advantages. These networks, on the other hand, have complicated and deeper network structure requiring high computational load. In this work, we present a shallow neural network structure that uses mainly two convolutional neural network (CNN) layers with few parameters, which requires low time and space complexity. In the experiments, we compare the proposed CNN models to three other neural network models with different depths applied in a mental arithmetic task using eye-closed state adapted for patients suffering from motor disorders and a decline in visual functions. Experimental results showed that this shallow CNN model can quickly learn the features from the data and achieve the highest classification accuracy of 90.68% in comparison of the 80% accuracy with conventional method and 67-73% with other shallow or deep neural network models.

Keywords: Physiological systems modeling - Multivariate signal processing, Physiological systems modeling - Signal processing in physiological systems
Abstract: Abstract — Auditory steady-state responses (ASSRs) enable hearing threshold estimation based on electrophysiological measurements and are widely used in clinical practice. Traditionally, ASSRs are recorded from a few electroencephalography (EEG) electrodes placed on the scalp. Ear-EEG is a method in which the EEG is recorded from electrodes placed within or around the ear and is thus more suitable for use in everyday life. Ear-EEG is typically recorded from multiple electrodes in order to enhance redundancy and robustness, but a pair of electrodes (so-called “best pair”) is usually chosen for the further analysis. Spatial filtering uses an optimized weighted combination of the electrodes, and is thus in general a better method for analysis of multichannel EEG. In this study we propose a new spatial filtering method based on solving a constrained optimization problem. Empirical evaluation based on ear-EEG recorded from nine subjects shows that the proposed spatial filtering method provides a significant increase in ASSR SNR as compared to the conventional “best pair” method. Clinical Relevance — ASSR can be estimated from ear-EEG recordings. Integrating ear-EEG into hearing aids would allow hearing aids to characterize hearing loss, and thereby adjust the audio processing accordingly.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Generally, those patients with dysarthria utter a distorted sound and the restrained intelligibility of a speech for both human and machine. To enhance the intelligibility of dysarthric speech, we applied a deep learning-based speech enhancement (SE) system in this task. Conventional SE approaches are used for shrinking noise components from the noise-corrupted input, and thus improve the sound quality and intelligibility simultaneously. In this study, we are focusing on reconstructing the severely distorted signal from the dysarthric speech for improving intelligibility. The proposed SE system prepares a convolutional neural network (CNN) model in the training phase, which is then used to process the dysarthric speech in the testing phase. During training, paired dysarthric normal speech utterances are required. We adopt a dynamic time warping technique to align the dysarthric–normal utterances. The gained training data are used to train a CNN-based SE model. The proposed SE system is evaluated on the Google automatic speech recognition (ASR) system and a subjective listening test. The results showed that the proposed method could notably enhance the recognition performance for more than 10% in each of ASR and human recognitions from the unprocessed dysarthric speech.

Keywords: Bioinformatics - Computational modeling and simulations in biology, physiology and medicine, Bioinformatics - High throughput –omics (genomics, proteomics, metabolomics, lipidomics, and metagenomics) data analytics for precision health
Abstract: Metabolism plays an important role in the pathogenesis of prostate cancer (PCa). Hence, we explored candidate metabolic-related genes attributed to biochemical relapse (BCR) of PCa. Gene expression profile and clinical parameters were downloaded from GSE70769 as “training set”. Using univariate Cox and LASSO-COX regression models, risk scores (RSs) were constructed. Kaplan-Meier (K-M) survival and time-dependent receiver operating characteristic (t-ROC) curves were employed. Univariate and multivariate Cox models were utilized to validate prognostic factors for biochemical relapse-free survival (BCRFS). Nomogram was plotted to facilitate clinical application. The dataset obtained from GSE70768 served as “validation set”. RSs were constructed by using 7 metabolic-related genes. RSs could significantly predict 1, 3, 5-year BCRFS (AUCs for training set: 0.810-0.836; AUC for validation set: 0.673-0.827). Nomograms could effectively predicted BCRFS (training set: C-index=0.831; validation set: C-index=0.737). RSs model is an independent prognostic factor for BCR, holding greater predictive value than traditional clinicopathological parameters.

Keywords: Bioinformatics - Computational modeling and simulations in biology, physiology and medicine, Health Informatics - Disease profiling and personalized treatment, Health Informatics - Computer-aided decision making
Abstract: Being diagnosed with breast cancer (BC) can be a traumatic experience for patients who may experience symptoms of depression. In order to facilitate the prevention of such symptoms, it is crucial to understand how and why depressive symptoms emerge and evolve for each individual, from diagnosis through treatment and recovery. In the present work, data from a multicentric study of 706 BC patients followed for 12 months are analyzed. First, a trajectory-based unsupervised clustering based on K-means is performed to capture the dynamic patterns of change in patients’ depressive symptoms after BC diagnosis and to identify distinct trajectory clusters. Then a supervised learning approach was employed to build a classification model of depression progression and to identify potential predictors. Patients were clustered into 4 groups: stable low, stable high, improving, and worsening depressive symptoms. In a nested cross-validation pipeline, the performance of the Support Vector Machine model for discriminating between “good” and “poor” progression was 0.78±0.05 in terms of AUC. Several psychological variables emerged as highly predictive of the evolution of depressive symptoms with the most important ones being negative affectivity and anxious preoccupation. The findings of the present study may help clinicians tailor individualized psychological interventions aiming at alleviating the burden of these symptoms in women with breast cancer and improving their overall well-being.

Keywords: Bioinformatics - Gene expression pattern recognition, Bioinformatics - Translational bioinformatics, General and theoretical informatics - Computational disease profiling
Abstract: Abstract— Comparative analyses utilizing publicly available big data have the potential to generate novel hypotheses and knowledge. However, this approach is underutilized in the realm of cancer research, particularly for prostate cancer. While the general progression of prostate cancer is now well understood, how individual cell types transition from healthy, to pre-cancerous, to cancerous cell types, remains to be further elucidated. To address this, we re-analyzed two publicly available single-cell RNA-seq datasets of prostate cancer and benign prostate hyperplasia cell types. The differential expression analysis of 15,505 epithelial cell profiles across 18,638 genes revealed 791 genes that were upregulated in prostate cancer epithelial cells. Here we report six markers that show significant upregulation in prostate cancer cells relative to BPH epithelial cells: HPN (5.62X), RAC3 (3.51X), CD24 (2.18X), HOXC6 (1.77X), AGR2 (1.71X), and IGFBP2 (1.28X). In particular, the significant differential expression of AGR2 further supports its clinical relevance in supplementing prostate-specific antigen screening for detecting prostate cancer. These findings have the potential to further advance our knowledge of genes governing the development of cancer in prostate epithelial cells.

Clinical Relevance— Our results establish the importance of 6 prostate cancer markers (HPN, RAC3, CD24, HOXC6, AGR2, and IGFBP3) in distinguishing between prostate cancer epithelial cells and benign prostate hyperplasia epithelial cells.

Keywords: Bioinformatics - High throughput –omics (genomics, proteomics, metabolomics, lipidomics, and metagenomics) data analytics for precision health
Abstract: Abstract— A meta-analysis study was conducted to compare high-throughput technologies in the classification of Adult-Onset Still’s Disease patients, using differentially expressed genes from independent profiling experiments. We exploited two publicly available datasets from the Gene Expression Omnibus and performed a separate differential expression analysis on each dataset to extract statistically important genes. We then mapped the genes of the two datasets and subsequently we employed well-established machine learning algorithms to evaluate the denoted genes as candidate biomarkers. Using next-generation sequencing data, we managed to achieve the maximum (100%) classification accuracy, sensitivity and specificity with the Gradient Boosting and the Random Forest classifiers, compared to the 83% of the DNA microarray data. Clinical Relevance— When biomarkers derived from one study are applied to the data of another, in many cases the results may diverge significantly. Here we establish that in cross-profiling meta-analysis approaches based on differential expression analysis, next-generation sequencing data provide more accurate results than microarray experiments in the classification of Adult-Onset Still’s Disease patients.

Keywords: Bioinformatics - Integration of multi-modality omic data, Bioinformatics - Computational systems biology, Bioinformatics - Gene expression pattern recognition
Abstract: Recent studies have confirmed the role of miRNA regulation of gene expression in oncogenesis for various cancers. In parallel, prior knowledge about relationships between miRNA and mRNA have been accumulated from biological experiments or statistical analyses. Improved identification of disease-associated miRNA-mRNA pairs may be achieved by incorporating prior knowledge into integrative genomic analyses. In this study we focus on 39 patients with hepatocellular carcinoma (HCC) and 25 patients with liver cirrhosis and use a flexible Bayesian two-step integrative method. We found 66 significant miRNA-mRNA pairs, several of which contain molecules that have previously been identified as potential biomarkers. These results demonstrate the utility of the proposed approach in providing a better understanding of relationships between different biological levels, thereby giving insights into the biological mechanisms underlying the diseases, while providing a better selection of biomarkers that may serve as diagnostic, prognostic, or therapeutic biomarker candidates.

Keywords: Bioinformatics - Sequencing alignment, assembly, and analysis, Bioinformatics - High throughput –omics (genomics, proteomics, metabolomics, lipidomics, and metagenomics) data analytics for precision health
Abstract: Myers bit-vector algorithm for approximate string matching (ASM) is a dynamic programming based approach that takes advantage of bit-parallel operations. It is one of the fastest algorithms to find the edit distance between two strings. In computational biology, ASM is used at various stages of the computational pipeline, including proteomics and genomics. The computationally intensive nature of the underlying algorithms for ASM operating on the large volume of data necessitates the acceleration of these algorithms. In this paper, we propose a novel ASM architecture based on Myers bit-vector algorithm for parallel searching of multiple query patterns in the biological databases. The proposed parallel architecture uses multiple processing engines and hardware/software codesign for an accelerated and energy-efficient design of ASM algorithm on hardware. In comparison with related literature, the proposed design achieves 22× better performance with a demonstrative energy efficiency of ~500×109 cell updates per joule.

Keywords: Physiological monitoring - Instrumentation, Physiological monitoring - Novel methods, Sensor systems and Instrumentation
Abstract: In recent years, wearable mid-activity electrical bioimpedance (EBI) sensing has been used to non-invasively track changes in edema and swelling levels within human joints. While the physiological origin of the changes in mid-activity EBI measurements have been demonstrated, EBI waveform patterns during activity have not been explored. In this work, we present a novel approach to extract waveform features from EBI measurements during gait to estimate the changes in vertical ground reaction forces (vGRF) corresponding to fatigue. Wearable EBI and vGRF data were measured from six healthy subjects during an asymmetric fatiguing protocol. For the exercised leg, the first peak of vGRF corresponding to the initial phase of simple support, decreased significantly and the loading rate increased significantly between the beginning and the end of the protocol. No significant change in these parameters were observed for the control leg. The first peak of vGRF and loading rate during the protocol (15 walking sessions) were correlated to the multi-frequency EBI features with mean Pearson’s r=0.81 and r=0.777, respectively. The results of this proof-of-concept study demonstrate the feasibility of estimating biomechanical parameters during activity with wearable EBI.

Keywords: Integrated sensor systems, Health monitoring applications, Sensor systems and Instrumentation
Abstract: Gait disturbances with falls are common among patients with Parkinson’s disease. Falls commonly occur from slips while walking on pathways with turns. Gait phases namely Loading Response and Terminal Stance are linked with forward and backward slips. Also, postural deformities (connected with knee joint angles) are debilitating symptoms of Parkinson’s patients and are related with falls. Here, we have focused on exploring the contribution of Loading Response and Terminal Stance to risk of fall along with the relevance of postural deformity (e.g., knee bending) while an individual walked overground on pathways (with 00 and 1800 turn) under dual task condition. For this, we have used a wearable device consisting of a pair of Sensored shoes and Knee Bending Angle Recorder Units. The device was used to compute Coefficient of Variation of knee bending angle during different gait phases as an indicator of one’s risk of fall that corroborated with clinical measure. Clinical Relevance- A study with age and gender matched healthy and Parkinson’s individuals indicated the importance of Loading Response and pathway turn while assessing risk of fall. This can serve as important pre-clinical input while designing intervention paradigms.

Keywords: Bio-electric sensors - Sensor systems, Integrated sensor systems, Bio-electric sensors - Sensing methods
Abstract: Diabetic peripheral neuropathy (DPN) affects a large proportion of people with diabetes, and early detection is essential to prevent further progression. Widespread clinical testing relies on simplicity and cost-effectiveness of examination. Early signs of DPN may be detected by assessing the sudomotor nerves, and sudomotor activity can be measured by bioimpedance. We present a prototype toe probe for DPN detection including sensors for measuring skin AC conductance, skin temperature and humidity. The prototype was tested on five participants with DPN and five healthy age-matched controls in a pilot study. Sudomotor sensor responses to a simple deep breathing test were very weak or absent in the DPN group, with all controls having larger responses than the DPN group. Evaporation was lower for the DPN group, and skin temperature was higher on average. For the same foot, the results for sudomotor responses were in agreement with sensory neurography amplitudes from the sural nerve whereas the monofilament test gave normal results for two of the DPN participants. If sufficient detection accuracy is confirmed in larger studies, the method may provide a simple and cost-effective tool to support clinical examination.

Keywords: Sensor systems and Instrumentation, Bio-electric sensors - Sensor systems
Abstract: The camptocormia angle has been established as a strong indicator for evaluating the progress of Parkinson's disease and the efficacy of therapeutical approaches. A wearable setup is proposed to measure the camptocormia angle with the perpendicular method using five inertial sensors. This study identifies suitable inertial measurement unit sensors for mobile long-term measurement. Moreover, a machine-learning approach is presented for segmenting the recorded data into periods with different dominant activities. An artificial neural network was the better classifier compared to a support vector machine to recognize certain common activities in patients with camptocormia. The artificial neural network's accuracy, sensitivity, and F1-score were 92.4 %, 82.9 %, and 82.1 %, respectively.

Keywords: Physiological monitoring - Modeling and analysis, Wearable sensor systems - User centered design and applications, Modeling and analysis
Abstract: Automatic lying posture tracking is an important factor in human health monitoring. The increasing popularity of the wrist-based trackers provides the means for unobtrusive, affordable long-term monitoring with minimized privacy concerns for the end-users and promising results in detecting the type of physical activity, step counting, and sleep quality assessment. However, there is limited research on development of accurate and efficient lying posture tracking models using wrist-based sensor. Our experiments demonstrate a major drop in the accuracy of the lying posture tracking using wrist-based accelerometer sensor due to the unpredictable noise from arbitrary wrist movements and rotations while sleeping. In this paper, we develop a deep transfer learning method that improves performance of lying posture tracking using noisy data in an arbitrary setting (wrist sensor data) by transferring the knowledge from an initial setting which contains both clean and noisy data. The proposed solution develops an optimal mapping model from the noisy data to the clean data in the initial setting using LSTM sequence regression, and reconstruct clean synthesized data in another setting where on noisy sensor data is available. This increases the lying posture tracking F1-Score by 24.9% for `left-wrist' and by 18.1% for `right-wrist' sensors comparing to the case without mapping.

Keywords: Sensor systems and Instrumentation, Health monitoring applications, Physiological monitoring - Novel methods
Abstract: Human pose estimation from monocular video is a rapidly advancing field that offers great promise to human movement science and rehabilitation. This potential is tempered by the smaller body of work ensuring the outputs are clinically meaningful and properly calibrated. Gait analysis, typically performed in a dedicated lab, produces precise measurements including kinematics and step timing. Using over 7000 monocular video from an instrumented gait analysis lab, we trained a neural network to map 3D joint trajectories and the height of individuals onto interpretable biomechanical outputs including gait cycle timing and sagittal plane joint kinematics and spatiotemporal trajectories. This task specific layer produces accurate estimates of the timing of foot contact and foot off events. After parsing the kinematic outputs into individual gait cycles, it also enables accurate cycle-by-cycle estimates of cadence, step time, double and single support time, walking speed and step length.

Keywords: Cardiovascular and respiratory signal processing - Cardiovascular signal processing, Cardiovascular regulation - Heart rate variability, Cardiovascular and respiratory signal processing - Heart Rate and Blood Pressure Variability
Abstract: More than a century has passed since Einthoven laid the foundation of modern electrocardiography and in recent years, driven by the advance of wearable and low budget devices, a sample accurate detection of R-peaks in noisy ECGsignals has become increasingly important. To accommodate these demands, we propose a new R-peak detection approach that builds upon the visibility graph transformation, which maps a discrete time series to a graph by expressing each sample as a node and assigning edges between intervisible samples. The proposed method takes advantage of the high connectivity of large, isolated values to weight the original signal so that Rpeaks are amplified while other signal components and noise are suppressed. A simple thresholding procedure, such as the widely used one by Pan and Tompkins, is then sufficient to accurately detect the R-peaks. The weights are computed for overlapping segments of equal size and the time complexity is shown to be linear in the number of segments. Finally, the method is benchmarked against existing methods using the same thresholding on a noisy and sample accurate database. The results illustrate the potential of the proposed method, which outperforms common detectors by a significant margin.

Keywords: Cardiovascular and respiratory signal processing - Non-linear cardiovascular or cardiorespiratory relations, Cardiovascular regulation - Autonomic nervous system, Cardiovascular and respiratory system modeling - Cerebrovascular models
Abstract: Quantification of the cardiorespiratory and cerebrovascular couplings is a relevant clinical issue given that their changes are considered signs of pathological status. The inherent nonlinearity of mechanisms underlying cardiorespiratory and cerebrovascular links requires nonlinear tools for their reliable evaluation. In the present study we compare two nonlinear methods for the assessment of coupling strength between two time series, namely cross-sample entropy (CSampEn) and k-nearest-neighbor cross-predictability (KNNCP). CSampEn uses a strategy that fixes the pattern length, while KNNCP optimizes the pattern length to maximize cross-predictability. CSampEn and KNNCP were applied to the beat-to-beat series of heart period (HP) and respiration (R) during a controlled breathing protocol with the aim at assessing cardiorespiratory coupling and to the beat-to-beat series of mean cerebral blood flow (MCBF) and mean arterial pressure (MAP) during an orthostatic stressor with the aim at evaluating cerebrovascular coupling. Although both the methods have the possibility to quantify the degree of HP-R and MCBF-MAP association, they exhibited different statistical power and even diverse trends in response to the considered physiological challenges. CSampEn and KNNCP are not interchangeable and should be utilized in association more than in alternative for the quantification of the HP-R and MCBF-MAP coupling strength.

Keywords: Cardiovascular and respiratory signal processing - Complexity in cardiovascular or respiratory signals, Cardiovascular regulation - Heart rate variability, Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: Several approaches for estimating complexity in physiological time series at various time scales have recently been developed, with a special focus on heart rate variability (HRV) series. While numerous multiscale complexity quantifiers have been investigated, a multiscale Kolmogorov-Sinai (K-S) entropy for the characterization of cardiovascular dynamics still has to be properly assessed. In this pilot study, we investigate the Algorithmic Information Content, which is calculated using an effective compression algorithm, to quantify multiscale partition-based K-S entropy on experimental HRV series. Data were gathered from publicly available datasets comprising longterm, unstructured recordings from 10 healthy subjects, as well as 10 patients with congestive heart failure (CHF) and 10 patients with atrial fibrillation. Results show that multiple time scales and domain partitions statistically discern healthy vs. pathological cardiovascular dynamics. We conclude that the proposed multiscale partition-based K-S entropy may constitute a viable tool for the complexity assessment of cardiovascular variability series.

Keywords: Pulmonary and critical care - Bioengineering applications in Intensive care
Abstract: While mechanical ventilation (MV) can lead to ventilator-induced diaphragmatic atrophy due to diaphragm inactivity, electrical phrenic nerve stimulation (PNS) can keep the diaphragm active and therefore prevent diaphragmatic weakness. To quantify the effectivity of PNS, an identification experiment during PNS is presented, and its data is used in Gaussian process regression (GPR) of the tidal volume based on the constant voltage amplitude of the stimulation pulses. The measurements were split into training data of variable size and test data for cross validation. For variable training sizes and different PNS settings, the GPR had a root mean square deviation (RMSD) between 0.39 and 0.91 mL/kg. An identification experiment as short as one and a half minutes was able to characteristically capture the relationship between tidal volume and voltage amplitude. The proposed method needs to be validated in further experiments.

Keywords: Cardiovascular regulation - Heart rate variability, Cardiovascular and respiratory system modeling - Gas exchange models, Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: During incremental exercise, two ventilatory thresholds (VT1, VT2) can normally be identified from gas exchange and ventilatory measurements, such as oxygen uptake, carbon dioxide production and ventilation. In this paper, we attempt to estimate the VT2 using HRV indices derived from a wearable electrocardiogram during a maximal exercise test. The exercise test is conducted on a treadmill that raises its speed by 0.5 km/h every minute. We have 42 measured exercise tests from 24 healthy male volunteers. Three experts determined the VT2 in each exercise test independently and we used principal component subspace reconstruction of their determinations to compute a collective VT2 for our machine learning model. The results demonstrate that the VT2 can be estimated from HRV using the proposed method with a reasonable performance during a maximal exercise test. In 28 out of 42 exercise tests, the HRV-derived threshold (HRVT) is within a minute (one phase) of the collective expert’s determination.

Keywords: Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: In this paper, a multiple linear regression model for estimating the Carotid-to-femoral pulse wave velocity (cf-PWV) from a single non-invasive peripheral pulse wave, namely blood pressure or photoplethysmography, is proposed. The training and testing datasets were extracted from in-silico, publicly available, pulse waves and hemodynamics data. The proposed model relies on a preprocessing and features extraction steps, which are performed using a semi-classical signal analysis (SCSA) method. The obtained results provide more evidence for the feasibility of machine learning and the SCSA method as a smart tool for the efficient assessment of the cf-PWV.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: Systems biology and systems neurophysiology in particular have recently emerged as powerful tools for a number of key applications in the biomedical sciences. Still, such models are often based on complex combinations of multiscale (and possibly multiphysics) strategies which require ad-hoc computational strategies and pose extremely high computational demands. Recent developments in the field of deep neural networks have demonstrated the possibility to formulate nonlinear, universal approximators to estimate solutions to highly nonlinear and complex problems with significant speed and accuracy advantages in comparison with traditional models. After synthetic data validation, we use so-called physically constrained neural netoworks (PINN) to simultaneusly solve the biologically plausible Hodgkin-Huxley model and infer its parameters and hidden time-courses from real data under both variable and constant current stimulation, demonstrating extremely low variability across spikes and faithful signal reconstruction. The parameter ranges we obtain are also compatible with prior knowledge. We demonstrate that detailed biological knowledge can be provided to a neural network making it able to fit complex dynamics over both simulated and real data.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Time-frequency analysis, Physiological systems modeling - Signal processing in physiological systems
Abstract: In recent years, as a treatment for mental disorders in addition to drug treatment, a non-drug treatment called chronotherapy has been attracting attention. However, the achievement of optimized chronotherapy for each subject's condition requires that the disturbance of the patient's circadian rhythm must be captured over a long duration. Therefore, it is necessary to develop biomarkers that are easy to measure, quantitative, and continuously measured. Complexity analysis of electroencephalograms revealed specific patterns related to circadian rhythms. However, such complexity analysis cannot capture variability in spatial patterns, although moment-to-moment temporal dynamic characteristics can be captured. Therefore, it is necessary to evaluate the dynamic characteristics of the interaction of neural activity throughout the brain. To evaluate the dynamic whole-brain interaction, we proposed a new microstate approach based on the instantaneous frequency distribution. In this context, we hypothesized that it would be possible to detect circadian rhythms using the microstate approach. In this study, to clarify the dynamic interactions of the entire neural network of the brain by circadian rhythms, we measured EEG data at day and night, and detected dynamic state transitions based on the instantaneous frequency distribution of the whole brain from EEG. The results showed the probability of transition among region-specific phase-leading states related to circadian rhythms. This finding might be widely utilized to detect circadian rhythms in healthy and pathological conditions.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Signal processing in simulation, Nonlinear dynamic analysis - Biomedical signals
Abstract: Astrocytes are recently considered active components in neural communication by modulating tripartite synaptic activity and the signaling mechanism facilitated by intercellular calcium wave (ICW) propagation. The heterogeneity in astrocytic connectivity produces diverse spatiotemporal signals equating to a diverse influence in synaptic communication. We developed a functional model of a neuron-astrocyte network consisting of tripartite synaptic interactions, gap-junction coupled astrocytic network, intra-, intercellular calcium diffusion, and varying topology to determine the effects of astrocytic connectivity to synaptic communication. The results suggest that the degree of astrocytic connectivity is vital in controlling the extrasynaptic glutamate to avoid disruption in synaptic communication.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification, Connectivity
Abstract: Archetypal analysis (AA) is a versatile data analysis method to cluster distinct features within a data set. Here, we demonstrate a framework showing the power of AA to spatio-temporally resolve events in calcium imaging, an imaging modality commonly used in neurobiology and neuroscience to capture neuronal communication patterns. After validation of our AA-based approach on synthetic data sets, we were able to characterize neuronal communication patterns in recorded calcium waves.

Keywords: Nonlinear dynamic analysis - Nonlinear filtering, Physiological systems modeling - Signal processing in physiological systems
Abstract: Monitoring the evolution of the Covid19 pandemic constitutes a critical step in sanitary policy design. Yet, the assessment of the pandemic intensity within the pandemic period remains a challenging task because of the limited quality of data made available by public health authorities (missing data, outliers and pseudoseasonalities, notably), that calls for cumbersome and ad-hoc preprocessing (denoising) prior to estimation. Recently, the estimation of the reproduction number, a measure of the pandemic intensity, was formulated as an inverse problem, combining data-model fidelity and space-time regularity constraints, solved by nonsmooth convex proximal minimizations. Though promising, that formulation lacks robustness against the limited quality of the Covid19 data and confidence assessment. The present work aims to address both limitations: First, it discusses solutions to produce a robust assessment of the pandemic intensity by accounting for the low quality of the data directly within the inverse problem formulation. Second, exploiting a Bayesian interpretation of the inverse problem formulation, it devises a Monte Carlo sampling strategy, tailored to a nonsmooth log-concave a posteriori distribution, to produce relevant credibility interval-based estimates for the Covid19 reproduction number. Applied to daily counts of new infections made publicly available by the Health Authorities for around 200 countries, the proposed procedures permit robust assessments of the time evolution of the Covid19 pandemic intensity, updated automatically and on a daily basis.

Keywords: Physiological systems modeling - Closed loop systems, Nonlinear dynamic analysis - Nonlinear filtering, Nonlinear dynamic analysis - Biomedical signals
Abstract: Currently, continuous glucose monitoring sensors are used in the artificial pancreas to monitor blood glucose levels. However, insulin and glucagon concentrations in different parts of the body cannot be measured in real-time, and determining body glucagon sensitivity is not feasible. Estimating these states provides more information about the current system status, facilitating improved decision-making by the model-based controller. In this regard, the aim of this paper is to design a nonlinear high-gain observer for a bi-hormonal artificial pancreas in the presence of measurement noises, model uncertainties, and disturbances. The model used in the observer is based on an existing intraperitoneal nonlinear animal model in the literature. This model is modified by assuming that insulin can directly transfer from the peritoneal cavity to the bloodstream. Based on a set of realistic assumptions, one model is considered after each hormone infusion, and two observers are separately designed. The model is divided into the insulin-phase and glucagon-phase models based on a set of realistic assumptions. Thereafter, two high-gain observers are designed separately for these phases contributing to estimating the non-measurable states. The observer error is proven to be locally uniformly ultimately bounded, and it is verified that any asymptotically stable control laws remain stable in the presence of the observer. The performance of the observers with different gains is evaluated for a scenario with multiple insulin and glucagon infusions. The proposed observer converges to a finite error, according to the results.

Clinical relevance-In Type 1 diabetic patients, the developed observer can be employed in a closed-loop artificial pancreas to improve the performance of model-based controllers. It estimates the key states, which are necessary for forecasting the body's response to insulin and glucagon boluses.

Keywords: Machine learning / Deep learning approaches, Image analysis and classification - Machine learning / Deep learning approaches, Image reconstruction and enhancement - Machine learning / Deep learning approaches
Abstract: The joint angular velocity during daily life exercises is an important clinical outcome for injury risk index, rehabilitation progress monitoring and athlete’s performance evaluation. Recently, wearable sensors have been widely used to monitor lower limb kinematics. However, these sensors are difficult and inconvenient to use in daily life. To mitigate these limitations, this study proposes a vision-based system for estimating lower limb joint kinematics using a deep convolution neural network with bi-directional long-short term memory and gated recurrent unit network. The normalized correlation coefficient, and the mean absolute error were computed between the ground truth obtained from the optical motion capture system and estimated joint angular velocities using proposed models. The estimated results show a highest correlation 0.93 in squat and 0.92 in walking on treadmill action. Furthermore, independent model for each joint angular velocity at the hip, knee, and ankle were analyzed and compared. Among the three joint angular velocities, knee joint has a best estimated accuracy (0.96 in squat and 0.96 in walking on the treadmill). The proposed models show higher estimation accuracy under both the lateral and the frontal view regardless of the camera positions and angles. This study proves the applicability of using sensor free vision-based system to monitor the lower limb kinematics during home workouts for healthcare and rehabilitation.

Keywords: PET and SPECT Imaging applications, Brain imaging and image analysis, PET and SPECT imaging
Abstract: Cardiac arrest (CA) is a fatal disease with high rates of neurological impairment. At present, targeted temperature management (TTM) is the only strategy with firm clinical evidence to prove its effectiveness. However, there is still controversy on the implementation of TTM, particularly on its depth, with a lack of elucidated underlying therapeutic mechanisms. Six Wistar rats were subjected to 8 min asphyxia-CA and randomly divided into TTM at 33° C (n=3) or 35° C groups (n=3). The spatiotemporal characteristics of cerebral glucose metabolism after CA were investigated by 18F-FDG microPET/CT. Myelin Basic Protein (MBP) immunofluorescence staining was used to assess acute injury and recovery of oligodendrocytes. Functional recovery was evaluated using the neurological deficit score (NDS). There was a significant improvement in functional recovery by NDS (p < 0.05) in the 33° C group compared with the 35° C group. Glucose metabolism of the 33° C group was higher than that of the 35° C group early after resuscitation (within 10 minutes). Immunofluorescence analysis showed that positive MBP signals in the cortex and hippocampus in the 33° C group were greater than in the 35° C group. In conclusion, compared to 35° C TTM, 33° C TTM changed the spatiotemporal characteristics of brain glucose metabolisms with improved neurological function, which may be through oligodendrocyte participation.

Keywords: PET and SPECT Imaging applications, Image analysis and classification - Machine learning / Deep learning approaches, Functional image analysis
Abstract: Positron emission tomography (PET) is able to reveal metabolic activity in a voxel-wise manner. PET analysis is commonly performed in a static manner by analyzing the standardized uptake value (SUV), obtained from plateau region of PET acquisitions. A dynamic PET acquisition can provide a map of the spatio-temporal concentration of the tracer in vivo, hence conveying information about radiotracer delivery to tissue, its interaction with the target and washout. Therefore, tissue specific biochemical properties are embedded in the shape of time activity curves (TACs), which are generally used for kinetic analysis. Conventionally TACs are employed along with information about blood plasma activity concentration i.e., the arterial input function (AIF), and specific compartmental models to obtain a full quantitative analysis of PET data. The main drawback of this approach is the need for invasive procedures requiring arterial blood sample collection during the whole PET scan. In this paper, we address the challenge of improving the PET diagnostic accuracy through an alternative approach based on the analysis of time signal intensity patterns. Specifically, we demonstrate the diagnostic potential of tissue TACs provided by a dynamic PET acquisition using various deep learning models. Our framework is shown to outperform the discriminative potential of classical SUV analysis, hence paving the way for a more accurate PET-based lesion discrimination without additional acquisition time or invasive procedures.

Keywords: Ultrasound imaging - High-frequency technology, Ultrasound imaging - Doppler, Image reconstruction and enhancement - Filtering
Abstract: Visualization of cutaneous micro-vasculatures is a determined approach in the diagnosis of skin vascular disorders. Clinically, high frequency ultrasound (HFUS) modalities have been used for cutaneous morphological and structural imaging, but visualization of micro-vessels has always been remained a daunting task. These tiny structures might be visualized by devising a highly sensitive Doppler technique for HFUS systems. In this study, we proposed an imaging framework using HFUS (30 MHz) ultrafast Doppler imaging along with SVD clutter filtering that is proficient in detection of micro-scale circulation. The performance of the devised framework was examined on a 200-micron flow phantom made of poly-vinyl alcohol under four different flow rates (56 – 18 ul/min) and visualized the micro-structure with averaged detected diameter of 93 – 170 μm. The results indicated that the devised framework has sufficient sensitivity and resolvability to visualize the micro-vasculatures in dermis layer of skin (depth ≤ 4 mm).

Keywords: Ultrasound imaging - High-frequency technology, Ultrasound imaging - Elastography, Multimodal imaging
Abstract: Despite advances in MRI, the detection and characterisation of lymph nodes in rectal cancer remains complex, especially when assessing the response to neo-adjuvant treatment. An alternative approach is functional imaging, previously shown to aid characterization of cancer tissues. We report proof-of-concept of the novel technique Contrast-Enhanced Magneto-Motive Ultrasound (CE-MMUS) to recover information relating to local perfusion and lymphatic drainage, and interrogate tissue mechanical properties through magnetically induced tissue deformations. The feasibility of the proposed application was explored using a combination of pre-clinical ultrasound imaging and finite element analysis. First, contrast enhanced ultrasound imaging on one wild type mouse recorded lymphatic drainage of magnetic microbubbles after bolus injection. Second, preliminary CE-MMUS data were acquired as a proof of concept. Third, the magneto-mechanical interactions of a magnetic microbubble with an elastic solid were simulated using finite element software. Accumulation of magnetic microbubbles in the inguinal lymph node was verified using contrast enhanced ultrasound, with peak enhancement occurring 3.7 s post-injection. Preliminary CE-MMUS indicates the presence of magnetic contrast agent in the lymph node. The finite element analysis explores how the magnetic force is transferred to motion of the solid, which depends on force, elasticity, and bubble radius, indicating an inverse relation between displacement and the latter two. Combining magnetic microbubbles with MMUS could harness the advantages of both techniques, to provide perfusion information, robust lymph node delineation and characterisation based on mechanical properties. Clinical Relevance— Robust detection and characterisation of lymph nodes could be aided by visualising lymphatic drainage of magnetic microbubbles using contrast enhanced ultrasound imaging and magneto-motion, which is dependent on tissue mechanical properties.

Keywords: Brain imaging and image analysis, Image analysis and classification - Machine learning / Deep learning approaches, Multivariate image analysis
Abstract: According to the 2016 World Health Organization(WHO) Classification scheme for gliomas, isocitrate dehydrogenase(IDH) is a very important basis for

diagnosis. There is a strong relationship between IDH mutation status and glioma prognosis. Therefore, it is important to predict the IDH mutation status for preoperatively treating glioma. In the past decade, there has been an increase in the use of machine learning, particularly deep learning, for medical diagnosis. To date, many methods using either deep learning or radiomics have been proposed for predicting glioma IDH mutation status. In this study, we proposed an intra- and inter-modality fusion model, which first fuses both Magnetic Resonance Imaging based (MRI-based) radiomics with deep learning features in

each modality (intra-modality fusion) and then the prediction results from each modality are fused by using an inter-modality regression model, to improve the IDH status prediction accuracy. The effectiveness of the proposed method is validated via our private glioma data set from the First Affiliated Hospital of Zhengzhou University (FHZU) in Zhengzhou, China. Our proposed method is superior to current state-of-the-art methods with an accuracy of 0.77, precision of 0.77, recall of 0.77, and F1 score of 0.77, thereby exhibiting an 8% increase in accuracy to predict the IDH mutation status for glioma treatment.

Keywords: Brain-computer/machine interface, Neural signals - Machine learning & Classification, Brain functional imaging - EEG
Abstract: Improving user performances is one of the major issues for Motor Imagery (MI) - based BCI control. MI-BCIs exploit the modulation of sensorimotor rhythms (SMR) over the motor and sensorimotor cortices to discriminate several mental states and enable user interaction. Such modulations are known as Event-Related Desynchronization (ERD) and Synchronization (ERS), coming from the mu (7-13 Hz) and beta (15-30 Hz) frequency bands. This kind of BCI opens up promising fields, particularly to control assistive technologies, for sport training or even for post-stroke motor rehabilitation. However, MI-BCIs remain barely used outside laboratories, notably due to their lack of robustness and usability (15 to 30% of users seem unable to gain control of an MI-BCI). One way to increase user performance would be to better understand the relationaships between user traits and ERD/ERS modulations underlying BCI performance. Therefore, in this article we analyzed how cerebral motor patterns underlying MI tasks (i.e., ERDs and ERSs) are modulated depending (i) on nature of the task (i.e., right-hand MI and left-hand MI), (ii) the session during which the task was performed (i.e., calibration or user training) and (iii) on the characteristics of the user (e.g., age, gender, manual activity, personality traits) on a large MI-BCI data base of N=70 participants. One of the originality of this study is to combine the investigation of human factors related to the user's traits and the neurophysiological ERD modulations during the MI task. Our study revealed for the first time, associations between ERD and both age, level of study, impression management and anxiety.

Keywords: Brain-computer/machine interface, Neural signal processing, Neural interfaces - Bioelectric sensors
Abstract: This study details the development of a novel, approx. £20 electroencephalogram (EEG)-based brain-computer interface (BCI) intended to offer a financially and operationally accessible device that can be deployed on a mass scale to facilitate education and public engagement in the domain of EEG sensing and neurotechnologies. Real-time decoding of steady-state visual evoked potentials (SSVEPs) is achieved using variations of the widely-used canonical correlation analysis (CCA) algorithm: multi-set CCA and generalised CCA. All BCI functionality is executed on board an inexpensive ESP32 mi- crocontroller. SSVEP decoding accuracy of 95.56 ± 3.74% with an ITR of 102 bits/min was achieved with modest calibration.

Keywords: Neural interfaces - Microelectrode technology, Neural interfaces - Implantable systems, Neural interfaces - Biomaterials
Abstract: Flexible polymer-based microelectrode arrays (MEAs) can reduce tissue inflammation and foreign body response and greatly prolong the lifetime of neural implants. However, standard and customized polymer devices are only accessible to limited groups. To better promote the development and application of polymer MEAs, we have launched the Polymer Implantable Electrode (PIE) Foundry and developed a 64-channel Parylene C-based MEA with generic electrodes layout that can be used to record from both cortical and sub-cortical regions in rodents. In addition, a practical dip-coating protocol for the insertion of the flexible standard Parylene MEA is developed.

Keywords: Brain-computer/machine interface, Neural signals - Machine learning & Classification
Abstract: A calibration procedure is required in motor imagery-based brain-computer interface (MI-BCI) to tune the system for new users. This procedure is time-consuming and prevents naive users from using the system immediately. Developing a subject-independent MI-BCI system to reduce the calibration phase is still challenging due to the subject-dependent characteristics of the MI signals. Many algorithms based on machine learning and deep learning have been developed to extract high-level features from the MI signals to improve the subject-to-subject generalization of a BCI system. However, these methods are based on supervised learning and extract features useful for discriminating various MI signals. Hence, these approaches cannot find the common underlying patterns in the MI signals and their generalization level is limited. This paper proposes a subject-independent MI-BCI based on a supervised autoencoder (SAE) to circumvent the calibration phase. The suggested framework is validated on dataset 2a from BCI competition IV. The simulation results show that our SISAE model outperforms the conventional and widely used BCI algorithms, common spatial and filter bank common spatial patterns, in terms of the mean Kappa value, in eight out of nine subjects.

Keywords: Brain-computer/machine interface
Abstract: Brain-Computer Interfaces (BCIs) based on Steady State Visually Evoked Potentials (SSVEPs) have proven effective and provide significant accuracy and information-transfer rates. This family of strategies, however, requires external devices that provide the frequency stimuli required by the technique. This limits the scenarios in which they can be applied, especially when compared to other BCI approaches. In this work, we have investigated the possibility of obtaining frequency responses in the EEG output based on the pure visual imagination of SSVEP-eliciting stimuli. Our results show that not only that EEG signals present frequency-specific peaks related to the frequency the user is focusing on, but also that promising classification accuracies can be achieved, paving the way for a robust and reliable visual imagery BCI modality.

Brain computer interfaces play a fundamental role in enhancing the quality of life of patients with severe motor impairments. Strategies based on purely imagined stimuli, like the one presented here, are particularly impacting, especially in the most severe cases.

Keywords: Neural signal processing, Neurological disorders - Stroke, Brain-computer/machine interface
Abstract: Low Frequency Brain Oscillations (LFOs) are brief periods of oscillatory activity in delta and lower theta band that appear at motor cortical areas before and around movement onset. It has been shown that LFO power decreases in post-stroke patients and re-emerges with motor functional recovery. To date, LFOs have not yet been explored during the motor execution (ME) and imagination (MI) of simple hand movements, often used in BCI-supported motor rehabilitation protocols post-stroke. This study aims at analyzing the LFOs during the ME and MI of the finger extension task in a sample of 10 healthy subjects and 2 stroke patients in subacute phase. The results showed that LFO power peaks occur in the preparatory phase of both ME and MI tasks on the sensorimotor channels in healthy subjects and their alterations in stroke patients.

Keywords: Brain imaging and image analysis, Image registration, segmentation, compression and visualization - Volume rendering, Magnetic resonance imaging - MR neuroimaging
Abstract: In this paper, we propose and validate an automatic pipeline for subcortical surface generation by making use of the average symmetrical surface distance (ASSD) loss in active shape modeling (ASM). A group of template surfaces are first generated via large deformation diffeomorphic metric mapping based surface deformation. ASM is then employed to obtain the mean shape and shape variation parameters of the template surfaces. To obtain the optimal shape variation parameters which best fit the target structure after acting upon the mean shape, a recently proposed derivative-free optimization method (the slow-fast learning method) is adopted. The ASSD loss, in addition to the typically utilized Dice similarity coefficient loss, is employed during the learning process to help enhance the boundary accuracy. We successfully validate the importance of the ASSD loss through ablation analysis. In addition, we show the effectiveness of the slow-fast learning method by comparing it with other state-of-the-art derivative-free optimization algorithms. Our proposed pipeline is found to be capable of yielding subcortical surfaces with high accuracy, correct anatomical topology, and sufficient smoothness.

Keywords: Brain imaging and image analysis, Image feature extraction, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Traditionally, the diagnosis of schizophrenia was based on the psychiatrist's introspective diagnosis through clinical stratification factors and score-scales, which led to heterogeneity and discrepancy in the symptoms and results. However, there are many studies trying to improve and assist in how its diagnosis could be performed. To objectively classify schizophrenia patients it is required to determine quantitative biomarkers of the disease. In this contribution we propose a method based on feature extraction both in magnetic resonance (MR) and Positron Emission Tomography (PET) imaging. A dataset of 34 participants (17 patients and 17 control subjects) were analyzed and 5 different brain regions were studied (frontal cortex, posterior cingulate cortex, temporal cortex, primary auditory cortex and thalamus). Following a radiomics approach, 43 texture features were extracted using five different statistical methods. These features were used for the training of the five different predictive models (Linear SVM, Gaussian SVM, Bagged Tree, KNN and Naïve Bayes). The precision results were obtained classifying schizophrenia both in MR images (89% Area Under the Curve (AUC) in the posterior cingulate cortex) and with PET images (82% AUC in the frontal cortex), being Linear SVM and Naïve Bayes the classification models with the highest predictive power.

Keywords: Brain imaging and image analysis, Image analysis and classification - Machine learning / Deep learning approaches, Optical imaging and microscopy - Neuroimaging
Abstract: As advances in microscopy imaging provide an ever clearer window into the human brain, accurate reconstruction of neural connectivity can yield valuable insight into the relationship between brain structure and function. However, human manual tracing is a slow and laborious task, and requires domain expertise. Automated methods are thus needed to enable rapid and accurate analysis at scale. In this paper, we explored deep neural networks for dense axon tracing and incorporated axon topological information into the loss function with a goal to improve the performance on both voxel-based segmentation and axon centerline detection. We evaluated three approaches using a modified 3D U-Net architecture trained on a mouse brain dataset imaged with light sheet microscopy and achieved a 10% increase in axon tracing accuracy over previous methods. Furthermore, the addition of centerline awareness in the loss function outperformed the baseline approach across all metrics, including a boost in Rand Index by 8%.

Keywords: Brain imaging and image analysis, PET and SPECT imaging
Abstract: Quantification of brain [18F]fluorodeoxyglucose ([18F]FDG) positron emission tomography (PET) data requires an input function. A noninvasive alternative to gold-standard arterial sampling is the image-derived input function (IDIF), typically extracted from the internal carotid arteries (ICAs), which are however difficult to segment and subjected to spillover effects. In this work, we evaluated the feasibility of extracting the IDIF from two different vascular sites, i.e., 1) common carotids (CCA) and 2) superior sagittal sinus (SSS), other than 3) ICA in a large group of glioma patients undergoing a dynamic [18F]FDG PET acquisition on a hybrid PET/MR scanner. Comparisons are drawn between the different IDIFs in terms of peak amplitude and shape, as well as between the estimates of fractional uptake rate (Ki) obtained from the different extraction sites in terms of a) grey/white matter average absolute values, b) ratio of grey-to-white matter, and c) spatial patterns for the hemisphere contralateral to the lesion.

Keywords: Brain imaging and image analysis, Functional image analysis, Magnetic resonance imaging - MR neuroimaging
Abstract: Neuropsychiatric disorders affect millions of people worldwide every year. Recent studies showed that the symptomatic overlaps across neuropsychiatric disorders mislead schizophrenia and bipolar disorder diagnosis. Additionally, recent studies claimed that schizoaffective disorder as a condition overlapped with both schizophrenia and bipolar disorder. Since symptomatic overlap among these disorders causes misdiagnosis, a need for neuroimaging biomarkers differentiating these disorders for a more accurate diagnosis is crucial. This study investigates dynamics functional network connectivity (dFNC) in the default mode network (DMN) of schizophrenia, bipolar, and schizoaffective disorder patients and compares them with their relative and healthy control. Additionally, it explored whether DMN dFNC features can predict the symptom severity of these neuropsychiatric disorders. Here, we found that dFNC features can differentiate schizophrenia from bipolar disorder. At the same time, we did not see a significant difference between schizoaffective with other conditions. Additionally, we found dFNC features can predict symptom severity of these three conditions.

Keywords: Brain imaging and image analysis, Image feature extraction, Magnetic resonance imaging - MR neuroimaging
Abstract: Brain functional connectivity has been shown to provide a type of fingerprint for adult subjects. However, most studies tend to focus on the connectivity strength rather than its stability across scans. In this study, we performed for the first time a large-scale analysis of within-individual stability of functional connectivity (FC) using 9071 children from the Adolescent Brain Cognitive Development database. Functional network connectivity (FNC) was extracted via a fully automated independent component analysis framework. We found that children's FNC is robust and stable with high similarity across scans and serves as a fingerprint that can identify an individual child from a large group. The robustness of this finding is supported by replicating the identification in the two-year follow-up session and between longitudinal sessions. More interestingly, we discovered that the within-individual FNC stability was predictive of cognitive performance and psychiatric problems in children, with higher FNC stability correlating with better cognitive performance and fewer dimensional psychopathology. The overall results indicate that the FNC of children also shows reliable within-individual stability, acting as a fingerprint for distinguishing participants, regardless of significant growth and development in the children's brain. FC stability can be a valuable imaging marker to predict early cognitive and psychiatric behaviors in children.

Keywords: Physiological systems modeling - Multivariate signal processing, Multivariate methods, Directionality
Abstract: Electroencephalography (EEG) microstates analysis provides a sequence of topographies representing the scalp-related electric field over time, and each microstate is synthetically represented by a symbol. Despite recent advances on functional brain-heart interplay (BHI) assessment, to our knowledge no methodology takes EEG microstates into account to relate the causal heartbeat dynamics. Moreover, standard BHI methods are tailored to a single EEG-channel analysis, neglecting the comprehensive information associated with a multichannel cluster or a whole-brain activity. To overcome these limitations, we devised a novel methodological framework for the assessment of functional BHI that exploits the symbolic representation of both EEG microstates and heart rate variability (HRV) series. Directional BHI quantification is then performed through Kullback-Leibler Divergence (KLD) and Transfer Entropy. The proposed methodology is here preliminarily tested on a dataset gathered from healthy subjects undergoing a resting state and a mental arithmetic task. Except for the KLD in the from-brain-to-heart direction, all other estimates showed significant differences between the two experimental conditions. We conclude that the proposed framework may promisingly provide novel insights on brain-heart phenomena through a whole-brain symbolic representation.

Keywords: Connectivity, Nonlinear dynamic analysis - Phase locking estimation, Coupling and synchronization - Coherence in biomedical signal processing
Abstract: Seizures frequently occur in paediatric emergency and critical care, with up to 74% being sub-clinical seizures making detection difficult. Delays in seizure detection and treatment worsen the neurological outcome of critically-ill patients. Gold-standard seizure detections using multi-channels electroencephalograms (EEG) require trained clinical physiologists to apply scalp electrodes and highly specialised neurologists to interpret and identify seizures. In this study, we extracted phase synchrony and cross-channel coherence amplitude across 4 and 8 pre-selected scalp EEG signals. Binary classification is used to determine whether the signal segment is seizure or non-seizure, and the predictions were compared against the gold-standard seizure onset markings. The application of the algorithm on a cohort of forty routinely collected EEGs from paediatric patients showed an average accuracy of 77.2 % and 76.5% using 4 and 8 channels, respectively.

Keywords: Coupling and synchronization - Nonlinear coupling, Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Physiological systems modeling - Signal processing in physiological systems
Abstract: Phase-amplitude coupling (PAC) based on the uniform phase empirical mode decomposition (UPEMD) is proposed to improve the accuracy of PAC assessment. The framework is applied to investigate the mechanism and improvement measure of gait disturbance for Parkinson’s disease (PD). Hβ modulation is suppressed at the time of contralateral heel strikes and rebounds when the contralateral foot rests on the ground and the ipsilateral foot is raised. Prominent PACs exist between δ and Lβ/Hβ activities. Auditory cue improves the gait; meanwhile, it enhances the Hβ modulation, and suppresses the δ-Lβ/Hβ PACs, which may rebound toward the before-cue stage afterward. Our findings suggest the proposed UPEMD-PAC is a useful framework in quantifying PAC with pre-determined frequencies, whereas the δ-Lβ/Hβ PACs in the subthalamic nucleus serve as potential biomarkers for gait disturbance in PD.

Keywords: Data mining and big data methods - Pattern recognition, Connectivity, Signal pattern classification
Abstract: Through source estimation, high-density electroencephalogram (EEG) signals at scalp level can be converted into signals at cerebral cortex level, which helps to measure cortical activity during anesthesia induced changes in consciousness level to explore the mechanism. In this research, the high-density EEG of propofol-induced consciousness states alterations in 20 healthy adults were converted into cortical signals of 68 regions of interest (ROI), after alpha bandpass filtering, the pairwise orthogonal power envelope connectivity (PEC) was calculated. Then, due to the number of PECs was huge, the least absolute shrinkage and selection operator (LASSO) was used to select as few PECs as possible as the indicators to distinguish baseline (BS) and moderate sedation (MD) states. The results show that most PECs that can be used as indicators are related to ROI related to default mode network (DMN). At the same time, changes of thalamocortical connectivity and frontal-parietal connectivity could be observed, similar to the neuroimaging method of directly measuring cerebral cortical activity. By extracting the PEC as a classifier to classify the BS and MD States, the accuracy could reach more than 70%. Therefore, this method can not only reflect the mechanism of cortical activity alterations induced by anesthetics, but also provide a new idea for monitoring the depth of anesthesia in the future.

Keywords: Data mining and big data methods - Machine learning and deep learning methods
Abstract: Multimodal brain networks characterize complex connectivities among different brain regions from both structural and functional aspects and provide a new means for mental disease analysis. Recently, Graph Neural Networks (GNNs) have become a de facto model for analyzing graph-structured data. However, how to employ GNNs to extract effective representations from brain networks in multiple modalities remains rarely explored. Moreover, as brain networks provide no initial node features, how to design informative node attributes and leverage edge weights for GNNs to learn is left unsolved. To this end, we develop a novel multiview GNN for multimodal brain networks. In particular, we treat each modality as a view for brain networks and employ contrastive learning for multimodal fusion. Then, we propose a GNN model which takes advantage of the message passing scheme by propagating messages based on degree statistics and brain region connectivities. Extensive experiments on two real-world disease datasets (HIV and Bipolar) demonstrate the effectiveness of our proposed method over state-of-the-art baselines.

Keywords: Signal pattern classification, Connectivity, Data mining and big data methods - Biosignal classification
Abstract: In this work, an attempt has been made to characterize arousal and valence emotional states using Electroencephalogram (EEG) signals and Phase lag index (PLI) based functional connectivity features. For this, EEG signals are considered from a publicly available DEAP database. Signals are decomposed into four frequency bands, namely theta (θ, 4–7 Hz), alpha (α, 8–12 Hz), beta (β, 13–30 Hz), and gamma (γ, 30-45 Hz). Two features, namely relative PSD and PLI, are calculated from each band of signals with Welch's periodogram. Four classifiers, namely Random Forest (RF), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and K-Nearest Neighbor (KNN), are employed to discriminate the emotional states. Results show that the proposed approach can differentiate emotional states using EEG signals. It is observed that there is strong functional connectivity in Fp1-O2 and Fp2-Pz in all emotional states for different frequency bands. SVM classifier yields the highest classification performance for arousal, and RF yields the highest performance for valence in the γ band. The combination of all features performs the best for the valence dimension. Thus, the proposed approach could be extended for classifying various emotional states in clinical settings.

Keywords: Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis, Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: In this study, the Multivariate Empirical Mode Decomposition (MEMD) is applied to multichannel EEG to obtain scale-aligned intrinsic mode functions (IMFs) as input features for emotion detection. The IMFs capture local signal variation related to emotion changes. Among the extracted IMFs, the high oscillatory ones are found to be significant for the intended task. The Marginal Hilbert spectrum (MHS) is computed from the selected IMFs. A 3D convolutional neural network (CNN) is adopted to perform emotion detection with spatial-temporal-spectral feature representations that are constructed by stacking the multi-channel MHS over consecutive signal segments. The proposed approach is evaluated on the publicly available DEAP database. On binary classification of valence and arousal level (high versus low), the attained accuracies are 89.25% and 86.23% respectively, which significantly outperform previously reported systems with 2D CNN and/or conventional temporal and spectral features.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Time-frequency analysis, Causality
Abstract: This work explores the potential utility of neural network classifiers for real-time classification of field-potential based biomarkers in next-generation responsive neuromodulation systems. Compared to classical filter-based classifiers, neural networks offer an ease of patient-specific parameter tuning, promising to reduce the burden of programming on clinicians. The paper explores a compact, feed-forward neural network architecture of only dozens of units for seizure-state classification in refractory epilepsy. The proposed classifier offers comparable accuracy to filter-classifiers on clinician-labeled data, while reducing detection latency. As a trade-off to classical methods, the paper focuses on keeping the complexity of the architecture minimal, to accommodate the on-board computational constraints of implantable pulse generator systems.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: In recent years, due to the fundamental role played by the central nervous system in emotion expression, electroencephalogram (EEG) signals have emerged as the most robust signals for use in emotion recognition and inference. Current emotion recognition methods mainly employ deep learning technology to learn the spatial or temporal representation of each channel, then obtain complementary information from different EEG channels by adopting a multi-modal fusion strategy. However, emotional expression is usually accompanied by the dynamic spatio-temporal evolution of functional connections in the brain. Therefore, the effective learning of more robust long-term dynamic representations for the brain’s functional connection networks is a key to improving the EEGbased emotion recognition system. To address these issues, we propose a brain network representation learning method that employs self-attention dynamic graph neural networks to obtain the spatial structure information and temporal evolution characteristics of brain networks. Experimental results on the AMIGOS dataset show that the proposed method is superior to the state-of-the-art methods.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Connectivity
Abstract: Mental calculations involve various areas of the brain. The frontal, parietal and temporal lobes of the left hemisphere have a principal role in the completion of this typology of tasks. Their level of activation varies based on the mathematical competence and attentiveness of the subject under examination and the perceived difficulty of the task. Recent literature often investigates patterns of cerebral activity through fMRI, which is an expensive technique. In this scenario, EEGs represent a more straightforward and cheaper way to collect information regarding brain activity. In this work, we propose an EEG based method to detect differences in the cerebral activation level of people characterized by different abilities in carrying out the same arithmetical task. Our approach consists in the extraction of the activation level of a given region starting from the EEG acquired during resting state and during the completion of a subtraction task. We then analyze these data through Functional Data Analysis, a statistical technique that allows operating on biomedical signals as if they were functions. The application of this technique allowed for the detection of distinct cerebral patterns among the two groups and, more specifically, highlighted the presence of higher levels of activation in the parietal lobe in the population characterized by a lower performance.

Keywords: Connectivity, Data mining and big data methods - Biosignal classification
Abstract: Psychogenic non-epileptic seizures (PNES) are attacks that resemble epilepsy but are not associated with epileptic brain activity and are regularly misdiagnosed. The current gold standard method of diagnosis is expensive and complex. Electroencephalogram (EEG) analysis with machine learning could improve this.

A k-nearest neighbours (kNN) and support vector machine (SVM) were used to classify EEG connectivity measures from 48 patients with PNES and 29 patients with epilepsy. The synchronisation method - correlation or coherence - and the binarisation threshold were defined through experimentation. Ten network parameters were extracted from the synchronisation matrix. The broad, delta, theta, alpha, beta, gamma, and combined 'all' frequency bands were compared along with three feature selection methods: the full feature set (no selection), light gradient boosting machine (LGBM) and k-Best.

Coherence was the highest performing synchronisation method and 0.6 was the best coherence threshold. The highest balanced accuracy was 89.74%, produced by combining all six frequency bands and selecting features with LGBM, classified by the SVM. This method returned a comparatively high accuracy but at a high computation cost. Future research should focus on identifying specific frequency bands and network parameters to reduce this cost.

Clinical relevance - This study found that EEG connectivity and machine learning methods can be used to differentiate PNES from epilepsy using interictal recordings to a high accuracy. Thus, this method could be an effective tool in assisting clinicians in PNES diagnosis without a video-EEG recording of a habitual seizure.

Keywords: Coupling and synchronization - Nonlinear coupling, Nonlinear dynamic analysis - Biomedical signals, Connectivity
Abstract: Alzheimer’s disease (AD) is a neurodegenerative disease known to affect brain functional connectivity (FC). Linear FC measures have been applied to study the differences in AD by splitting neurophysiological signals such as electroencephalography (EEG) recordings into discrete frequency bands and analysing them in isolation. We address this limitation by quantifying cross-frequency FC in addition to the traditional within-band approach. Cross-bispectrum, a higher-order spectral analysis, is used to measure the nonlinear FC and is compared with the cross-spectrum, which only measures the linear FC within bands. Each frequency coupling is then used to construct an FC network, which is in turn vectorised and used to train a classifier. We show that fusing features from networks improves classification accuracy. Although both within-frequency and cross-frequency networks can be used to predict AD with high accuracy, our results show that bispectrum-based FC outperforms cross-spectrum suggesting an important role of cross-frequency FC.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Nonstationary analysis and modeling
Abstract: The use of brain-computer interface (BCI) technology has emerged as a promising rehabilitation approach for patients with motor function and motor-related disorders. BCIs provide an augmentative communication platform for controlling advanced assistive robots such as a lower-limb exoskeleton. Brain recordings collected by an electroencephalography (EEG) system have been employed in the BCI platform to command the exoskeleton. To date, the literature on this topic is limited to the prediction of gait intention and gait variations from EEG signals. This study, however, aims to predict the anticipated gait direction using a stream of EEG signals collected from the brain cortex. Three healthy participants (age range: 29-31, 2 female) were recruited. While wearing the EEG device, the participants were instructed to initiate gait movement toward the direction of the arrow triggers (pointing forward, backward, left, or right) being shown on a screen with a blank white background. Collected EEG data was then epoched around the trigger timepoints. These epochs were then converted to the time-frequency domain using event-related synchronization (ERS) and event-related desynchronization (ERD) methods. Finally, the classification pipeline was constructed using logistic regression (LR), support vector machine (SVM), and convolutional neural network (CNN). A ten-fold cross-validation scheme was used to evaluate the classification performance. The results revealed that the CNN classifier outperforms the other two classifiers with an accuracy of 0.75.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Deep learning techniques are increasingly used for decision-making in health applications, however, these can easily be manipulated by adversarial examples across different clinical domains. Their security and privacy vulnerabilities raise concerns about the practical deployment of these systems. The number and variety of adversarial attacks grow continuously, making it difficult for mitigation approaches to provide effective solutions. Current mitigation techniques often rely on expensive re-training procedures as new attacks emerge. In this paper, we propose a novel adversarial mitigation technique for biosignal classification tasks. Our approach is based on recent findings interpreting early exit neural networks as an ensemble of weight sharing sub-networks. Our experiments on state-of-the-art deep learning models show that early exit ensembles can provide robustness generalizable to various white box and universal adversarial attacks. The approach increases the accuracy of vulnerable deep learning models up to 60 percentage points, while providing adversarial mitigation comparable to adversarial training. This is achieved without previous exposure to the adversarial perturbation or the computational burden of re-training.

Keywords: Physiological systems modeling - Signal processing in simulation, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Magnetic resonance spectroscopy (MRS) is a noninvasive method that enables the analysis and quantification of brain metabolites, which provide useful information about the neuro-biological substrates of brain function. Lactate plays a pivotal role in the diagnosis of various brain diseases. However, accurate lactate quantification is generally difficult to achieve due to the presence of large lipid peaks resonating at a similar spectral position. To overcome this problem several techniques have been proposed. However, most of them suffer from lactate signal loss or poor lipid peak removal. In this paper, a novel method for lipid suppression for MRS signal is proposed. The method combines a semi-classical signal analysis method and a bidirectional long short term memory technique. The method is validated using simulated data that mimics real MRS data.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Patient outcome and risk analysis, Principal component analysis
Abstract: Sepsis is one of the leading causes of death in ICU and its timely recognition and management are of primary importance. Resuscitation from hypotension in patients with sepsis is one of the first challenges that require fluid and/or vasopressor administrations. Unfortunately, clinical guidelines provide only indications of the strategy that should be adopted in this critical population but personalized strategies are still missing.

In this study, we propose a comparative analysis of reinforcement learning applications on ICU data collected in the electronic health records and publicly available within the MIMIC-III database. The ultimate goal of the study is to estimate the optimal fluid and vasopressor administrations.

Results show that, after the use of principal component analysis for reducing feature space dimensionality, model performances increased, thus suggesting that additional preprocessing strategies might be used for both reducing the computational cost and refining model performances.

Clinical relevance - In a context where clinical guidelines are not able to provide the best treatment strategies at a patient level, reinforcement learning applications trained on retrospectively collected data may be used for developing models able to suggest to clinicians the optimal dosage of fluids and/or vasopressors in order to improve 90-day patients' survival.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals, Nonlinear dynamic analysis - Nonlinear filtering
Abstract: Automatic motion artifact (MA) removal in electrodermal activity (EDA) signals is a major challenge because of the aperiodic and irregular characteristics of EDA. Given the lack of a suitable MA removal algorithm, a substantial amount of EDA data is typically discarded, especially during ambulatory monitoring. Current methods for MA removal in EDA are feasible when data are corrupted with low magnitude artifacts. In this study, we propose a more data-driven deep convolutional autoencoder (DCAE) for automated motion artifact removal in EDA signals. The DCAE was trained using several publicly available datasets. We used both Gaussian white noise (GWN) and real-life induced MA data records collected in a laboratory setting to corrupt the clean EDA signals. We compared the performance of our DCAE network with three state-of-the-art methods using the performance metrics the signal-to-noise ratio (SNR) improvement (SNR_imp), and the mean squared error (MSE). The proposed DCAE provided significantly higher SNR_imp and lower MSE compared to three other methods for both synthetically and real-life induced MA. While the work is preliminary, this work illustrates a promising approach which can potentially be used to remove many different types of MA

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Systems identification
Abstract: Glucose prediction is used in diabetes self-management as it allows to take suitable actions for proper glycemic regulation of the patient. The aim of this work is the short-term personalized glucose prediction in patients with Type 1 diabetes mellitus (T1DM). In this scope, we compared two different models, an autoregressive moving average (ARMA) model and a long short-term memory (LSTM) model for different prediction horizons. The comparison of two models was performed using the evaluation metrics of root mean square error (RMSE) and mean absolute error (MAE). The models were trained and tested in 29 real patients. The results shown that the LSTM model had better performance than ARMA with RMSE 3.13, 6.41 and 8.81 mg/dL and MAE 1.98, 5.06 and 6.47 mg/dL for 5-, 15- and 30-minutes prediction horizon.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Time-frequency and time-scale analysis - Empirical mode decomposition in biosignal analysis
Abstract: Adequate patients'data have always been critical for disease assessment. However, large amounts of patient data are often difficult to collect, especially when patients are required to complete a series of assessment movements. For example, assessing the hand motor function of stroke patients or Parkinson’s patients requires patients to complete a series of evaluation movements, and it is often difficult for patients to complete each group of actions multiple times, resulting in a small amount of data. To solve the problem of insufficient data quantity, this study proposes a data augmentation method based on empirical mode decomposition and integrated long short-term memory neural network (EMD-ILSTM). The method mainly consists of two parts: one is to decompose the raw signal by the method of EMD, and the other is to use LSTM for data augmentation of the decomposed signal. Then, the method is tested on the public dataset named Ninaweb, and the test results show that the classification accuracy can be improved by 5.2% by using the augmented data for classification tasks. Finally, clinical trials are conducted to verify that after dimensionality reduction, the augmented data and raw data have smaller intra-class distances and larger inter-class distances, indicating that data augmentation is effective.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Resting-state fMRI is commonly used for diagnosing Autism Spectrum Disorder (ASD) by using network-based functional connectivity. It has been shown that ASD is associated with brain regions and their inter-connections. However, discriminating based on connectivity patterns among imaging data of the control population and that of ASD patients’ brains is a non-trivial task. In order to tackle said classification task, we propose a novel deep learning architecture (MHATC) consisting of multi-head attention and temporal consolidation modules for classifying an individual as a patient of ASD. The devised architecture results from an in-depth analysis of the limitations of current deep neural network solutions for similar applications. Our approach is not only robust but computationally efficient, which can allow its adoption in a variety of other research and clinical settings.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: The primary somatosensory cortex (S1) is a region often targeted for input via somatosensory neuroprosthesis as tactile and proprioception are represented in S1. How this information is represented is an ongoing area of research. Neural signals are high-dimensional, making accurate models for decoding a significant challenge. Artificial neural networks (ANNs) have proven efficient at classification tasks in multiple fields. Moreover, ANNs allow for transfer learning, which exploits feature extraction that was trained on a large and more general dataset than may be available for a particular problem. In this work, convolutional neural networks (CNN), used for image recognition, were fine-tuned with somatosensory cortical recordings during experiments with naturalistic touch stimuli. We created a highly accurate (> 96% correct) classifier for cutaneous stimulation locations as part of a somatosensory neuroprosthesis pipeline. Here we present the classifier results.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Nonlinear dynamic analysis - Biomedical signals
Abstract: Hypoxemia, characterized by low blood oxygen levels is pervasive in preterm infants and is associated with development of multiple adverse cardiovascular morbidities. In clinical practice, it is often quantified using frequency, pattern and time spent in it. A predictive tool of hypoxemia occurrence will aid clinicians in risk stratifying infant oxygenation patterns and improving personalized care. As a first step towards this goal in characterizing the underlying temporal processes, we studied inter-hypoxemia interval distributions in preterm infants on oxygen supplementation. We derived regression relationships of characterizing parameters of the distributions with gestational age and birth weight of infants. The modeling and goodness of fit tests of pooled and individual inter-hypoxemia intervals indicated that the inverse Gaussian and Birnbaum Saunders distributions fit well over short time scales and the lognormal at longer time scales. Information from distribution modeling may provide insights into hypoxemia recurrence times and be helpful in developing models to predict severe hypoxemic events that may be translated to personalized care in clinical settings.

Keywords: Causality, Directionality, Connectivity
Abstract: Productivity, stress, and work engagement play important roles in corporate health. In this study, we have investigated, on the basis of survey data, interactive-effect relationships among productivity, stress, and work engagement. Survey results obtained from 301 samples self-report questionnaires (including the WHO Health and Work Performance Questionnaire: HPQ; the Perceived Stress Scale: PSS; and the Utrecht Work Engagement Scale: UWES) were analyzed using mediation analysis. Results suggest that the interactive positive effects of productivity and work engagement were roughly equal, and that stress decreased both productivity and work engagement. Revealing the relationships among productivity, stress, and work engagement contributes to the efforts of occupational health physicians and of workers in human resource departments trying to plan effective and preferential interventions in order to improve employee working conditions.

Keywords: Causality, Physiological systems modeling - Signal processing in physiological systems, Neural networks and support vector machines in biosignal processing and classification
Abstract: Four different Granger causality-based methods -one linear and three nonlinear (Granger Causality, Kernel Granger Causality, large-scale Nonlinear Granger Causality, and Neural Network Granger Causality) were used for assessment and causal-based quantification of the respiratory sinus arrythmia (RSA) in the group of pediatric cardiac patients, based on the single-lead ECG and impedance pneumography signals (the latter as the tidal volume curve equivalent). Each method was able to detect the dependency (in terms of causal inference) between respiratory and cardiac signals. The correlations between quantified RSA and the demographic parameters were also studied, but the results differ for each method.

Keywords: Coupling and synchronization - Nonlinear synchronization, Nonlinear dynamic analysis - Phase locking estimation, Signal pattern classification
Abstract: Cardiorespiratory Phase Synchronization (CRPS) is the manifestation of the non-linear coupling between the cardiac and the respiratory systems, different to the Respiratory Sinus Arrythmia (RSA). This takes place when the heartbeats occur at the same relative phase of the breathing, during a succession of respiratory cycles. In this study, we investigated the CRPS in 45 elderly patients admitted to the semi-critical unit of a hospital. The patients were classified according to their respiratory state as non-Periodic Breathing (nPB), Periodic Breathing (PB) and Cheyne-Stokes Respiration (CSR). The phase synchrogram between the electrocardiographic and respiratory signals was computed using the Hilbert transform technique. A continuous measure of the CRPS was obtained from the synchrogram, and was characterized by the average duration of synchronized epochs (AvgDurSync), the percentage of synchronized time (%Sync), the number of synchronized epochs (NumSync), and the frequency ratio between the cardiac and respiratory oscillators (FreqRat). These measures were studied using two different thresholds (0.1 and 0.05) for the amplitude of the synchronization and a minimum duration threshold of 10s. According to the results, the AvgDurSync and %Sync had a decreasing trend in patients with breathing periodicity. In addition, CSR patients presented the lowest values AvgDurSync and %Sync. Therefore, the CRPS method could be a useful tool for characterizing periodic respiratory patterns in elderly patients, which might be related to chronic heart failure.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Time-frequency and time-scale analysis - Wavelets, Physiological systems modeling - Signal processing in physiological systems
Abstract: Atrial fibrillation (AF) is the most common cardiac arrhythmia and is associated with significant morbidity and mortality. Owing to the advances in sensor technology and the emergence of wearable devices that enable daily self-monitoring, ECG signal processing methods for the automatic detection of AF are more pertinent than ever. In this paper, we investigate the use of wavelet higher-order statistics (WHOS) for feature extraction and differentiation between normal sinus rhythm and AF. The proposed approach captures the evolution of the WHOS dynamics and quantifies the changes in the time-varying characteristics of the frequency couplings caused by AF. Results obtained from the statistical analysis of a dataset of 5834 single-lead ECG recordings, reveal 46/50 statistically significant features and provide insight into the complexity of the evolution of the ECG non-linearities during AF.

Keywords: Nonlinear dynamic analysis - Biomedical signals
Abstract: Despite advances in prenatal health care, neonatal sepsis remains a major cause of neonatal mortality. Early diagnosis and adequate treatment are essential to reduce morbidity and mortality related to this disease. In this paper, we propose a new method to detect neonatal sepsis based on heart rate (HR) complexity measures (entropy and compression indices) that takes into consideration neonatal gestational age. First, the percentile curves were computed for all the complexity indices using data from 118 control neonates. Eight indices were computed: the sample entropy (SampEn) and three indices to quantify the multiscale entropy (MSE) curve -- the sum, the slope, and the product of the previous two -- and the compression ratio (CR), using the bzip2 compressor, as well as the same three indices but related to the multiscale compression (MSC) curve. Then, the corresponding percentile was estimated for 23 sepsis neonates. Results show a significant decrease in the entropy indices SampEn and MSEsum and in the MSCslope a day before the detection of sepsis by the clinicians. The indices CR and MSCsum increased before the antibiotic take. These results imply that sepsis causes a random, uncorrelated pattern on the HR signal. Future studies should include a bigger data set to calculate a compound index comprising information of other physiological signals.

Keywords: Nonlinear dynamic analysis - Biomedical signals, Multivariate methods, Physiological systems modeling - Multivariate signal processing
Abstract: Functional near infrared spectroscopy (fNIRS) is a modality that can measure shallow cortical brain signals and also contains pulsatile oscillations that originate from heartbeat dynamics. In particular, while fNIRS slow waves (0 Hz to 0.6 Hz) refer to the standard hemodynamic signal, fast-wave (0.8 Hz to 3 Hz) fNIRS signals refer to cardiac oscillations. Using a cognitive stress experiment paradigm with mental arithmetic, the aim of this study was to assess differences in cortical activity when using slow-wave or fast- wave fNIRS signals. Furthermore, we aimed to see whether fNIRS fast and slow waves provide different information to discriminate mental arithmetic tasks from baseline. We used data from 10 healthy subjects from an open dataset performing mental arithmetic tasks and assessed fNIRS signals using mean values in the time domain, as well as complexity estimates including sample, fuzzy, and distribution entropy. A searchlight representational similarity analysis with pairwise t-test group analysis was performed to compare the representational dissim- ilarity matrices of each searchlight center. We found significant representational differences between fNIRS fast and slow waves for all complexity estimates, at different brain regions. On the other hand, no statistical differences were observed for mean values. We conclude that entropy analysis of fNIRS data may be more sensitive than traditional methods like mean analysis at detecting the additional information provided by fast-wave signals for discriminating mental arithmetic tasks and warrants further research.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Signal pattern classification
Abstract: Rhythmic bioelectrical ‘slow waves’ are a key regulatory mechanism underpinning digestion. The pyloric sphincter separates the independent slow wave and contractile behaviour of the stomach and small intestine, while also regulating gastric emptying. In this study, we develop and validate anatomically-specific electrode cradles and analysis techniques in pigs, to map in vivo slow wave activation across this critical pylorus region for the first time. 3D printed electrode cradles were developed from reconstructions of magnetic resonance images, to accurately capture anatomical geometry. A low-pass Savitzky-Golay filter with an equivalent cut-off frequency of ~2 Hz was chosen as the optimal filter for analysis of both gastric and intestinal slow waves. Preliminary results showed that slow waves in the terminal antrum occurred with a frequency of (2.81 ± 0.55) cycles per minute (cpm), velocity of (5.04 ± 0.29) mm s^-1 , and amplitude of (1.38 ± 0.37) mV, before terminating at a zone of quiescence at the pylorus that was (41.22 ± 7.4) mm wide. The proximal duodenal pacemaker was identified in several recordings, where it initiated slow waves at a frequency of (18.1 ± 0.80) cpm, velocity of (11.3 ± 2.4) mm s^-1, and amplitude of (0.376 ± 0.027) mV. This work enables quantitative definitions of numerous physiological features of the in vivo pylorus region, including the quiescent zone size and duodenal pacemaker location. This work establishes a novel method for in vivo measurement of bioelectrical slow wave activity of the pyloric region, which is a key target for physiological investigation and clinical intervention. In the future, the new methods developed here may be able to inform diagnosis and/or treatment of gastric motility diseases that affect the pylorus.

Keywords: Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Changes in emotional state, such as anxiety, have a significant impact on behavior and mental health. However, the detection of anxiety in individuals requires trained specialists to administer specialized assessments, which often take a significant amount of time and resources. Thus, there is a significant need for objective and real-time anxiety detection methods to aid clinical practice. Recent advances in Adaptive Mixture Independent Component Analysis (AMICA) have demonstrated the ability to detect changes in emotional states using electroencephalographic (EEG) data. However, given that several hours may be need to identify the different models, alternative methods must be sought for future brain-computer-interface applications. This study examines the feasibility of a machine learning classifier using frequency domain features of EEG data to classify individual 500 ms samples of EEG data into different cortical states, as established by multi-model AMICA labels. Using a random forest classifier with 12 input features from EEG data to predict cortical states yielded a 75% accuracy in binary classification. Based on these findings, this work may provide a foundation for real-time anxiety state detection and classification.

Keywords: Physiological systems modeling - Multivariate signal processing, Physiological systems modeling - Signal processing in simulation, Signal pattern classification
Abstract: This project seeks to assess the use of multivariate auto-regressive (MAR) models to create forecasts of continuous vital parameters in hospitalized patients. A total 20 hours continuous (1/60Hz) heart rate and respiration rate from 8 postoperative patients, where used to fit a centered MAR model for forecasting in windows of 15 minutes. The model were fitted using Markov Chain Monte Carlo sampling, and the model were evaluated on data from 5 new patients. The results show an average RMSE in the forecast window of 11.4(SD: 7.30) beats per minute for heart rate and 3.3(SD:1.3) breaths per minute for respiration rate.

Keywords: Signal pattern classification
Abstract: Automatic discrimination of clean and deformed segments of capnogram signals is an essential requisite in capnogram-based respiratory assessment. However, improving the performance of this classification task remains challenging, particularly in terms of specificity and sensitivity. The goal of this paper is to address this issue by proposing a cooperative classification approach rather than relying solely on a single classifier. The presented method’s main advantage is the vote participation of four distinct classifiers that affects the reliability of the final classification decision. MATLAB simulation was run on a dataset consisting of 200 15-seconds capnogram segments, 100 of which are clean and 100 are deformed. The results revealed a trade-off between the achieved specificity and sensitivity by adjusting the strictness of voting. Being highly strict in the sense of classifying a capnogram segment as clean if and only if all voting classifiers agreed on deciding so, provided specificity and sensitivity of 94% and 81%, respectively. On the contrary, lowering the strictness of voting by considering only one positive vote is sufficient to eventually classify the query capnogram segment as non-deformed gave specificity and sensitivity of 74% and 94%, respectively.

Keywords: Signal pattern classification, Physiological systems modeling - Signal processing in physiological systems
Abstract: Transcranial magnetic stimulation co-registered with electroencephalographic (TMS-EEG) has previously proven a helpful tool in the study of Alzheimer’s disease (AD). In this work, we investigate the use of TMS-evoked EEG responses to classify AD patients from healthy controls (HC). By using a dataset containing 17AD and 17HC, we extract various time domain features from individual TMS responses and average them over a low, medium and high density EEG electrode set. Within a leave-one-subject-out validation scenario, the best classification performance for AD vs. HC was obtained using a high-density electrode with a Random Forest classifier. The accuracy, sensitivity and specificity were of 92.7%, 96.58% and 88.82% respectively.

Keywords: Signal pattern classification, Physiological systems modeling - Signal processing in physiological systems
Abstract: Transcranial Magnetic Stimulation (TMS) combined with EEG recordings (TMS-EEG) has shown great potential in the study of the brain and in particular of Alzheimer’s Disease (AD). In this study, we propose an automatic method of determining the duration of TMS induced perturbation of the EEG signal as a potential metric reflecting the brain’s functional alterations. A preliminary study is conducted in patients with Alzheimer’s disease (AD). Three metrics for characterizing the strength and duration of TMS-evoked EEG (TEP) activity are proposed and their potential in identifying AD patients from healthy controls was investigated. A dataset of TMS-EEG recordings from 17 AD and 17 healthy controls (HC) was used in our analysis. A Random Forest classification algorithm was trained on the extracted TEP metrics and its performance is evaluated in a leave-one-subject-out cross-validation. The created model showed promising results in identifying AD patients from HC with an accuracy, sensitivity and specificity of 69.32%, 72.23% and 66.41%, respectively.

Keywords: Signal pattern classification
Abstract: In this paper, the paradigm of the intention of speed changes from EEG signals with Riemannian classifiers methods is studied in 10 subjects. In addition, the best frequency band and how different electrode configurations affect the accuracy of the model are analyzed. In the prediction of the intention to change speed, results of 68.6% were obtained, in the one of only Increase, results of 64.41% were obtained, and in the one of only Decrease, results of 71.5% were obtained.

Keywords: Signal pattern classification, Physiological systems modeling - Closed loop systems
Abstract: Affective brain-computer interfaces are a fast growing area of research. Accurate estimation of emotional states from physiological signals is of great interest to the fields of psychology and human-computer interaction. The DEAP dataset is one of the most popular datasets for emotional classification. In this study we generated heat maps from spectral data within the neurological signals found in the DEAP dataset. To account for the class imbalance within this dataset, we then discarded images belonging to the larger class. We used these images to fine-tune several Big Transfer neural networks for binary classification of arousal, valence, and dominance affective states. Our best classifier was able to achieve greater than 98% accuracy and 99% balanced accuracy in all three classification tasks. We also investigated the effects of this balancing method on our classifiers.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: Affective states play an important role in human behavior and decision-making. In recent years, several affective brain-computer interface (aBCI) studies have focused on developing an emotion classifier based on elicited emotions within the user. However, it is difficult to achieve consistency in elicited emotions across populations, which can lead to dataset imbalances. The experimental design presented in this paper seeks to avoid consistency issues by asking the participant to classify the emotion portrayed in images of facial expressions, rather than their own emotions. Priming is also a common technique used in psychology studies that is known to influence emotional perception. To improve participant accuracy, we investigated matching and mis-matched word priming for the facial expression images. Electro-encephalogram (EEG) data were used to generate images fed into a classifier based on the Big Transfer model, BiT-M R101x1. The primed images resulted in higher classification accuracy overall. Further, by building different classifier models for both mis-matched primed images and matching primed images, we were able to achieve classification accuracies above 90%. We also provided the classifier with the true labels of the photographs instead of the labels generated by the participants and achieved similar results. The experimental paradigm of measuring brain activity during the emotional classification of another individual provides consistently high, balanced classification accuracies.

Keywords: Signal pattern classification
Abstract: Dysphagia causes aspiration symptoms and can trigger aspiration pneumonia, poor nutritional status, etc. To address these risks, it is important to properly evaluate dysphagia and link it with treatment and training. However, current dysphagia evaluation methods cannot assess a swallowing function equivalent to that in daily life, owing to the examination method and environment. In this study, we analyzed bio-signal features to realize a system that can detect aspiration symptoms in daily life. Focusing on the neck electrical impedance, swallowing sounds, and a surface electromyogram of the suprahyoid muscles, we created a swallowing-measurement device and analyzed the bio-signals of the throat movement during swallowing. By measuring the swallowing of dysphagic patients, we investigated the characteristic differences, depending on the presence or absence of aspiration symptoms. The analysis results suggest that there were differences in each bio-signal depending on the presence or absence of aspiration symptoms, and these bio-signals could detect aspiration symptoms.

Keywords: Physiological systems modeling - Signal processing in physiological systems
Abstract: Electrodermal activity (EDA), which tracks sweat gland activity as a proxy for sympathetic activation, has the potential to be a biomarker of physiological and psychological changes in the clinic. To show this, in this study, we demonstrate that the tonic component of EDA responds consistently and robustly during induction of anesthesia in the operating room in 8 subjects during surgery. This response is seen bilaterally. The response shows a significant increase in EDA in anticipation of induction and then a gradual decrease in response to the administration of medication, which agrees with both the expected psychological effects of stress and anxiety and the physiological effects of anesthetic medication on sweat glands. The results also show a slightly faster response to drug in the arm directly receiving the medication intravenously compared to the opposite, though the magnitude of the effect evens out over time.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Pattern recognition
Abstract: Weaning from mechanical ventilation in the intensive care unit is a complex and relevant clinical problem. Prolonged mechanical ventilation leads to a variety of medical complications that increase hospital stay and costs, in addition to contributing the morbidity and mortality, affecting long-term quality of life. This work presents a methodology to establish the optimal moment of extubation of a patient connected to a mechanical ventilator, submitted to the T-Tube test. 133 patients are analyzed, classified into two groups: successful group (94 patients) and failed group (39 patients). The behaviour of the respiratory function is characterized through the mean, standard deviation, kurtosis, skewness, interquartile range and coefficient of interval of the respiratory flow time series. To classify these patients, neural networks (NN) and support vector machines (SVM) classifier are used, considering time intervals of the 450s, 600s and 900s. According to the results, the best classification is obtained using the SVM.

Clinical Relevance — The paper determines the optimal moment for weaning a patient connected to a mechanical ventilator using machine learning techniques.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Image segmentation, Optical imaging and microscopy - Microscopy
Abstract: Imaging mass cytometry (IMC) is a new advancement in tissue imaging that is quickly gaining wider usage since its recent launch. It improves upon current tissue imaging methods by allowing for a significantly higher number of proteins to be imaged at once on a single tissue slide. For most analyses of IMC data, determining the phenotype of each cell is a crucial step. Current methods of phenotyping require sufficient biological knowledge regarding the protein expression profile of the various cell types. Here, we develop a deep convolutional autoencoder-classifier to automate the cell phenotyping process into four basic cell types. Biopsy tissue from bladder cancer patients is used to evaluate the efficacy of the classification. The model is evaluated and validated through feature importance, confirming that the significant features are biologically relevant. Our results demonstrate the potential of deep learning to automate the task of cell phenotyping for high-dimensional IMC data.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, CT imaging applications
Abstract: Synthetic medical images have an important role to play in developing data-driven medical image processing systems. Using a relatively small amount of patient data to train generative models that can produce an abundance of additional samples could bridge the gap towards big-data in niche medical domains. These generative models are evaluated in terms of the synthetic data they generate using the Visual Turing Test (VTT), Fréchet Inception Distance (FID), and other metrics. However, these are generally interpreted at the group level, and do not measure the artificiality of individual synthetic images. The present study attempts to address the challenge of automatically identifying artificial images that are obviously-artificial-looking, which may be necessary for filtering out poorly constructed synthetic images that might otherwise deteriorate the performance of assimilating systems. Synthetic computed tomography (CT) images from a progressively-grown generative adversarial network (PGGAN) were evaluated with a VTT and their image embeddings were analyzed for correlation with artificiality. Images categorized as obviously-artificial (≥0.7 probability of being rated as fake) were classified using a battery of algorithms. The top-performing classifier, a support vector machine, exhibited accuracy of 75.5%, sensitivity of 0.743, and specificity of 0.769. This is an encouraging result that suggests a potential approach for validating synthetic medical image datasets.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Image feature extraction, Ultrasound imaging - Vascular imaging
Abstract: Accurate measurement of blood vessel diameter on ultrasonic images is important in many vascular exams. In one of them, volumetric blood flow measurements, the volume flow rate is calculated by multiplying the time-averaged velocity with the cross-sectional area of the vessel (using diameter measured from B-mode images). Computation of lumen diameter is also vital for planning surgical procedures like carotid artery stenting and endarterectomy. More recently, several automated vessel diameter estimation methods employing deep learning have been proposed. In this paper, we propose a novel single-step automated deep learning-based vessel diameter estimation technique developed on B-mode images. Longitudinal images of the human common carotid artery were acquired by trained vascular sonographers in human subjects using a linear array probe. Ground truth measurements were obtained by a human expert to validate the proposed technique. 504 images (with augmentation) were divided into training, validation, and test sets. Three pre-trained deep learning networks were used for training, and the lumen diameter was predicted in a hold-out test set. The Mean Absolute Deviation (MAD) and Root Mean Square Error (RMSE) ranged from 0.22-0.65 mm and 0.32-0.82 mm, respectively, for the three networks. Furthermore, 5-fold cross-validation resulted in MAD and RMSE of 0.36±0.1 mm and 0.513±0.15 mm, respectively. Clinical Relevance— The results demonstrate that the technology can potentially be embedded in commercial scanners to make the workflow in vascular ultrasound more efficient.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Quality degradation (QD) is common in the fundus images collected from the clinical environment. Although diagnosis models based on convolutional neural networks (CNN) have been extensively used to interpret retinal fundus images, their performances under QD have not been assessed. To understand the effects of QD on the performance of CNN-based diagnosis model, a systematical study is proposed in this paper. In our study, the QD of fundus images is controlled by independently or simultaneously importing quantified interferences (e.g., image blurring, retinal artifacts, and light transmission disturbance). And the effects of diabetic retinopathy (DR) grading systems are thus analyzed according to the diagnosis performances on the degraded images. With images degraded by quantified interferences, several CNN-based DR grading models (e.g., AlexNet, SqueezeNet, VGG, DenseNet, and ResNet) are evaluated. The experiments demonstrate that image blurring causes a significant decrease in performance, while the impacts from light transmission disturbance and retinal artifacts are relatively slight. Superior performances are achieved by VGG, DenseNet, and ResNet in the absence of image degradation, and their robustness is presented under the controlled degradation.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Rigid-body image registration, Image reconstruction and enhancement - Image synthesis
Abstract: Craniosynostosis is a condition associated with the premature fusion of skull sutures affecting infants. 3D photogrammetric scans are a promising alternative to computed tomography scans in cases of single suture or nonsyndromic synostosis for diagnostic imaging, but oftentimes diagnosis is not automated and relies on additional cephalometric measurements and the experience of the surgeon.

We propose an alternative representation of the infant's head shape created from 3D photogrammetric surface scans as 2D distance maps. Those 2D distance maps rely on ray casting to extract distances from a center point to the head surface, arranging them into a 2D image grid. We use the distance map for an original CNN-based classification approach, which is evaluated on a publicly available synthetic dataset for benchmarking and also tested on clinical data.

Qualitative differences of different head shapes can be observed in the distance maps. The CNN-based classifier achieves accuracies of 100% on the publicly available synthetic dataset and 98.86% on the clinical test set.

Our distance map approach demonstrates the diagnostic value of 3D photogrammetry and the possibility of automatic, CNN-based diagnosis. Future steps include the improvement of the mapping method and testing the CNN on more pathologies.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Image feature extraction, Image classification
Abstract: Non-small cell lung cancer (NSCLC) is a malignant tumor with high morbidity and mortality, with a high recurrence rate after surgery, which directly affects the life and health of patients. Recently, many studies are based on Computed Tomography (CT) images. They are cheap but have low accuracy. In contrast, the use of gene expression data to predict the recurrence of NSCLC has high accuracy. However, the acquisition of gene data is expensive and invasive, and cannot meet the recurrence prediction requirement of all patients. In this paper, we proposed a low-cost, high-accuracy residual multilayer perceptrons (ResMLP) recurrence prediction method. First, several proposed ResMLP modules are applied to construct a deep regression estimation model. Then, we build a mapping function of mixed features (handcrafted features and deep features) and gene data via this model. Finally, the recurrence prediction task is realized, by utilizing the gene estimation data obtained from the regression model to learn the information representation related to recurrence. The experimental results show that the proposed method has strong generalization ability and can reach 86.38% prediction accuracy.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Image classification, Image segmentation
Abstract: Malignant transformation of gastric ulcer can result in gastric cancer, hence an accurate gastric ulcer classification method is of vital importance. Despite marvelous progress has been achieved in recent years, there are still many challenges in diagnosis of gastric ulcer. In this paper, we propose a mechanism to mimic gastroenterologist’s behaviours based on deep learning techniques, by integrating the segmented malignancy suspicious masks with gastroscopic images for gastric ulcer classification, which instructs the model to focus on the area where symptoms occur for gastric ulcer diagnosis. Specifically, a U-Net-type deep neural network is built to segment the suspicious pathological regions from gastroscopic images, then the segmented regions are treated as an attention channel of gastroscopic images for the gastric ulcer classification by a ResNet-type deep neural network. Experiments on a real gastroscopic dataset with 900+ patient cases demonstrate that our proposed approach achieves much better performance for gastric ulcer diagnosis, compared with standard method with only gastroscopic images.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Image feature extraction, Image classification
Abstract: An efficient face detector could be very helpful to point out possible neurological dysfunctions such as seizure events in Neonatal Intensive Care Units. However, its development is still challenging because large public datasets of newborns’ faces are missing. Over the years several studies introduced semi-automatic approaches. This study proposes a fully automated face detector for newborns in Neonatal Intensive Care Units, based on the Aggregate Channel Feature algorithm. The developed method is tested on a dataset of video recordings from 42 full-term newborns collected at the Neuro-physiopathology and Neonatology Clinical Units, AOU Careggi, Firenze, Italy. The proposed system showed promising results, giving (mean ± standard error): log-Average Miss Rate = 0.47 ± 0.05 and Average Precision Recall = 0.61 ± 0.05. Moreover, achieved results highlighted interesting differences between newborns without seizures, newborns with electro-clinical seizures, and newborns with electrographic-only seizures. For both metrics statistically significant differences were found between patients with electro-clinical seizures and the other two groups.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Novel imaging modalities
Abstract: The incidence of burn injuries is higher in low and middle-income countries, and particularly in remote areas where the access to specialized burn assessment, care and recovery is limited. Given the high costs associated with one of the most used techniques to evaluate the severity of a burn, namely laser Doppler imaging (LDI), an alternative approach could be beneficial for remote locations. This study proposes a novel approach to estimate the LDI from digital images of a burn. The approach is a pixel-wise regression model based on convolutional neural networks. To minimize the dependency on the conditions in which the images are taken, the effect of two image normalization techniques is also studied. Results indicate that the model performs satisfactorily on average, presenting low mean absolute and squared errors and high structural similarity index. While no significant differences are found when changing the normalization of the images, the performance is affected by their quality. This suggests that changes in the intensity of the images do not alter the relevant information about the wound, whereas changes in brightness, contrast and sharpness do.

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Machine learning / Deep learning approaches, Optical imaging and microscopy - Fluorescence microscopy
Abstract: Gene sequencing technology is a tool which greatly impacts modern biology and medicine. The next-generation sequencing (NGS) lies at the heart of gene sequencing for its massively increasing throughput, but it is difficult to analyze the large quantities of fluorescent images with high accuracy because the fluorescent signals are weak with varying noise signals, and current designs are limited on accuracy and speed. In this paper, we proposed a novel deep learning based gene sequencing pipeline with semi-automatic labelling method. The obtained results are promising, especially on the high-density data, as the BaseFormer surpasses the traditional methods in terms of cluster quality (Q30: 88%), throughput (16.5% better), and with similar and low error rate (down to 0.137% on average, best at 0.068% on high-density data).

Keywords: Image analysis and classification - Machine learning / Deep learning approaches, Brain imaging and image analysis, Magnetic resonance imaging - Other organs
Abstract: Intracranial arteries are critical blood vessels that supply the brain with oxygenated blood. Intracranial artery labels provide valuable guidance and navigation to numerous clinical applications and disease diagnoses. Various machine learning algorithms have been carried out for automation in the anatomical labeling of cerebral arteries. However, the task remains challenging because of the high complexity and variations of intracranial arteries. This study investigates a novel graph convolutional neural network with deep feature fusion for cerebral artery labeling. We introduce stacked graph convolutions in an encoder-core-decoder architecture, extracting high-level representations from graph nodes and their neighbors. Furthermore, we efficiently aggregate intermediate features from different hierarchies to enhance the proposed model’s representation capability and labeling performance. We perform extensive experiments on public datasets, in which the results prove the superiority of our approach over baselines by a clear margin.

Keywords: Image classification, Magnetic resonance imaging - Other organs
Abstract: Computer-aided diagnosis (CAD) of prostate cancer (PCa) using multi-parametric magnetic resonance imaging (mp-MRI) has recently gained great research interest. In this work, a fully automatic CAD pipeline of PCa using mp-MRI data is presented. In order to fully explore the mp-MRI data, we systematically investigate three multi-modal medical image fusion strategies in convolutional neural networks, namely input-level fusion, feature-level fusion, and decision-level fusion. Extensive experiments are conducted on two datasets with different PCa-related diagnostic tasks. We identify a pipeline that works relatively the best for both diagnostic tasks, two important components of which are stacking three adjacent slices as the input and performing decision-level fusion with specific loss weights.

Keywords: Image classification, Machine learning / Deep learning approaches, Magnetic resonance imaging - Other organs
Abstract: Early detection of precancerous cysts or neoplasms, i.e., Intraductal Papillary Mucosal Neoplasms (IPMN), in pancreas is a challenging and complex task, and it may lead to a more favourable outcome. Once detected, grading IPMNs accurately is also necessary, since low-risk IPMNs can be under surveillance program, while high-risk IPMNs have to be surgically resected before they turn into cancer. Current standards (Fukuoka and others) for IPMN classification show significant intra- and inter-operator variability, beside being error-prone, making a proper diagnosis unreliable. The established progress in artificial intelligence, through the deep learning paradigm, may provide a key tool for an effective support to medical decision for pancreatic cancer. In this work, we follow this trend, by proposing a novel AI-based IPMN classifier that leverages the recent success of transformer networks in generalizing across a wide variety of tasks, including vision ones. We specifically show that our transformer-based model exploits pre-training better than standard convolutional neural networks, thus supporting the sought architectural universalism of transformers in vision, including the medical image domain and it allows for a better interpretation of the obtained results.

Keywords: Image classification, Ultrasound imaging - Breast
Abstract: Medical ultrasound (US) imaging has become a prominent modality for breast cancer imaging due to its ease of use, low cost, and safety. In the past decade, convolutional neural networks (CNNs) have emerged as the method of choice in vision applications and have shown excellent potential in the automatic classification of US images. Despite their success, their restricted local receptive field limits their ability to learn global context information. Recently, Vision Transformer (ViT) designs, based on self-attention between image patches, have shown great potential to be an alternative to CNNs. In this study, for the first time, we utilize ViT to classify breast US images using different augmentation strategies. We also adopted a weighted cross-entropy loss function since breast ultrasound datasets are often imbalanced. The results are provided as classification accuracy and Area Under the Curve (AUC) metrics, and the performance is compared with the SOTA CNNs. The results indicate that the ViT models have comparable efficiency with or even better than the CNNs in the classification of US breast images.

Keywords: Image classification, Magnetic resonance imaging - MR neuroimaging, Machine learning / Deep learning approaches
Abstract: Abstract—Deep Learning approaches are powerful tools in a great variety of classification tasks. However, they are limitedly accepted or trusted in clinical frameworks due to their typical “black box” outline: their architecture is well-known, but processes employed in classification are often inaccessible to humans. With this work, we explored the problem of “Explainable AI” (XAI) in Alzheimer’s disease (AD) classification tasks. Data from a neuroimaging cohort (n = 251 from OASIS-3) of early-stage AD dementia and healthy controls (HC) were analysed. The MR scans were initially fed to a pre-trained DL model, which achieved good performance on the test set (AUC: 0.82, TPR: 0.78, TNR: 0.81). Results were then investigated by means of an XAI approach (Occlusion Sensitivity method) that provided measures of relevance (RV) as outcome. We compared RV values obtained within healthy tissues with those underlying white matter hyperintensity (WMH) lesions. The analysis was conducted on 4 different groups of data, obtained by stratifying correct and misclassified images according to the health condition of participants (AD/HC). Results highlighted that the DL model found favourable leveraging lesioned brain areas for AD identification. A statistically significant difference (p<0.01) between WMH and healthy tissue contributions was indeed observed for AD recognition, differently from the HC case (p=0.27). Clinical Relevance—This study, though preliminary, suggested that DL models might be trained to use known clinical information and reinforced the role of WMHs as neuroimaging biomarker for AD dementia. The outlined findings have a significant clinical relevance as they prepare the ground for a progressive increase in the level of trust laid in DL approaches.

Keywords: Image feature extraction, Image classification, Optical imaging and microscopy - Microscopy
Abstract: Human spermatozoa must swim through the female reproductive tract, where they undergo a series of biochemical and biophysical reactions called capacitation, a necessary step to fertilize the egg. Capacitation promotes changes in the motility pattern. Historically, a two-dimensional analysis has been used to classify sperm motility and clinical fertilization studies. Nevertheless, in a natural environment sperm motility is three-dimensional (3D). Imaging flagella of freely swimming sperm is a difficult task due to their high beating frequency of up to 25 Hz. Very recent studies have described several sperm flagellum 3D beating features (curvature, torsion, asymmetries, etc.). However, up to date, the 3D motility pattern of hyperactivated spermatozoa has not been characterized. The main difficulty in classifying these patterns in 3D is the lack of a ground truth reference since differences in flagellar beat patterns are very difficult to assess visually. Moreover, only around 10-20% of induced to capacitate spermatozoa are truly capacitated, i.e., hyperactivated. We used an image acquisition system that can acquire, segment, and track spermatozoa flagella in 3D+t. In this work, we propose an original three-dimensional feature vector formed by ellipses describing the envelope of the 3D+t spatio-temporal flagellar sperm motility patterns. These features allowed compressing an unlabeled 3D+t dataset to separate hyperactivated cells from others (capacitated from non-capacitated cells) using unsupervised hierarchical clustering. Preliminary results show three main clusters of flagellar motility patterns. The first principal component of these 3D flagella measurements correlated with 2D OpenCASA head determinations as a first approach to validate the unsupervised classification, showing a reasonable correlation coefficient near to 0.7.

Keywords: Image feature extraction, Magnetic resonance imaging - Other organs, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Osteoarthritis is one of the most disabling diseases in developed countries. Its etiology is not firmly established, and the diagnosis is made by observing radiographs, assigning a degree of severity based on the information displayed. For this reason, the diagnosis is usually late and determined by the subjectivity of the doctor, which implies a restriction of the treatment. Magnetic resonance imaging (MRI) has allowed us to see in greater detail the alterations produced in soft joint structures. In this work, biomarkers for an early diagnosis of knee osteoarthritis have been developed by means of textures analysis on MRI. For this purpose, 50 subjects underwent T1-weighted MR image acquisitions: 25 controls and 25 diagnosed with knee osteoarthritis between grades I and III. Six regions were segmented on these images, corresponding to the femorotibial cartilage, femoral condyles, and tibial plateau. 43 textures were extracted for each region of interest (ROI) employing 5 statistical methods and 5 different predictive models were trained and compared. In addition, a study of the thickness of the cartilage was carried out to make a comparison with the texture analysis. The best result has been obtained using a K-nearest neighbor model with the combination of 33 textures (maximum value of AUC = 0.7684). Furthermore, in the analysis of the cartilage thickness, no statistically significant differences were found. Finally, it is concluded that the texture analysis has great potential for the diagnosis of knee osteoarthritis.

Keywords: Image feature extraction, Regularized image Reconstruction, Image reconstruction and enhancement - Machine learning / Deep learning approaches
Abstract: Effectively modeling and quantifying behavior is essential for our understanding of the brain. Modeling behavior across different subjects in a unified manner remains a significant challenge in the field of behavioral quantification, which necessitates partitioning the behavioral data into features that are common across subjects, and others that are distinct to each subject. We build on a semi-supervised approach to partition the subspace adequately known as a Partitioned Subspace Variational AutoEncoder (PS-VAE), and propose a novel regularization based on the Cauchy-Schwarz divergence to model the distinct features across subjects. Our model, called the Cauchy-Schwarz regularized Partitioned Subspace Variational AutoEncoder (CS-PS-VAE), successfully models continuously varying differences in behavior, and models distinct features of the behavioral videos across subjects in an unsupervised manner. This method is also successful at uncovering the relationships between recorded neural data and the ensuing behavior.

Keywords: Image classification, Image feature extraction, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Screening programs for sight-threatening diseases rely on the grading of a large number of digital retinal images. As automatic image grading technology evolves, there emerges a need to provide a rigorous definition of image quality with reference to the grading task. In this work, on two subsets of the CORD database of clinically gradable and matching non-gradable digital retinal images, a feature set based on statistical and on task-specific morphological features has been identified. A machine learning technique has then been demonstrated to classify the images as per their clinical gradeability, offering a proxy for a rigorous definition of image quality.

Keywords: Image enhancement, Image reconstruction and enhancement - Tomographic reconstruction
Abstract: Breast cancer is one of the most diagnosed forms of cancer among women worldwide. However, the survival rate is very high when the tumor is diagnosed early. The search for diagnostic techniques increasingly able to detect lesions of the order of a few millimeters and to overcome the limitations of current diagnostic techniques (e.g., the X-ray mammography, currently used as standard for screening campaigns) is always active. Among the main emerging techniques, microwave and millimeter-wave imaging systems have been proposed, using either radar or tomographic approaches. In this paper, a novel dual-step millimeter-wave imaging which combines the advantages of tomographic and radar approaches is proposed. The goal of this work is to reconstruct the dielectric profile of suspicious regions by exploiting the morphological information from the radar maps as a priori information within quantitative tomographic techniques. Promising preliminary dielectric reconstruction results against simulated data are shown in both single- and dual-target scenarios, in which high-density healthy and tumor tissues are present. The reconstruction results were compared to the dielectric characteristics of human breast ex-vivo tissues used in the simulated models. The proposed dual-step approach is able to accurately reconstruct the real part of the complex permittivity of the targets, allowing to distinguish their nature also in the most challenging case represented by the co-presence of high-density healthy tissues and a malignant lesion, thus paving the way for a deeper investigation of this approach in experimental scenarios.

Keywords: Machine learning / Deep learning approaches, Optical imaging and microscopy - Fluorescence microscopy, Image registration, segmentation, compression and visualization - Volume rendering
Abstract: Cell segmentation at a single cell resolution is required to provide insights for basic biology and application study. However, there are issues of low signal-to-noise ratio, weak fluorescence response, and insufficient resolution along the image stacking direction in 3D confocal images (volume). It has been difficult to segment out single cells from close or contacted cells in a cell volume using image processing methods or together with geometric processing methods. Recently, 3D deep learning methods have been used to avoid tedious parameter settings in the image and geometric processing, but still not easy to segment out close or contacted single cells. This paper proposes a 2D U-net to segment cell regions in high accuracy and computing performance. Better 3D cell images and single cell segmentation for close or contacted cells are achieved by combining a 3D U-net to detect the centers of single cells in the volume.

Keywords: Machine learning / Deep learning approaches, Image analysis and classification - Digital Pathology
Abstract: Chronic Venous Disorders (CVD) of lower limbs are one of the most prevalent medical conditions, affecting 35% of adults in Europe and North America. The early diagnosis of CVD is critical, however, the diagnosis relies on a visual recognition of the various venous disorders which is time-consuming and dependent on the physician's expertise. Thus, automatic strategies for the classification of the CVD severity are claimed. This paper proposed an automatic ensemble-based strategy of Deep Convolutional Neural Networks (DCNN) for the classification of CVDs severity from medical images. First, a clinical dataset containing 1376 photographs of patients with CVD of 5 different levels of severity was constructed. Then, the constructed dataset was randomly split into training, testing, and validation datasets. Subsequently, a set of DCNN were individually applied to the images for classification. Finally, instead of a traditional voting ensemble strategy, extracted feature vectors from each DCNN were concatenated and fed into a new ensemble optimization network. Experiments showed that the proposed strategy achieved a classification with 93.8%, 93.4%, 92.4% of accuracy, precision, and recall, respectively. Moreover, compared to the traditional ensemble strategy, improvement in the accuracy of ~2% was registered. The proposed strategy showed to be accurate and robust for the diagnosis of CVD severity from medical images. Nevertheless, further research using an extensive clinical database is required to validate the potential of this strategy.

Keywords: Machine learning / Deep learning approaches, Image segmentation, CT imaging
Abstract: Domain adaptation has become an important topic because the trained neural networks from the source domain generally perform poorly in the target domain due to domain shifts, especially for cross-modality medical images. In this work, we present a new unsupervised domain adaptation approach called Multi-Stage GAN (MSGAN) to tackle the problem of domain shift for CT and MRI segmentation tasks. We adopt the multi-stage strategy in parallel to avoid information loss and transfer rough styles on low-resolution feature maps to the detailed textures on high-resolution feature maps. In detail, the style layers map the learnt style codes from the Gaussian noise to the input features in order to synthesize images with different styles. We validate the proposed method for cross-modality medical image segmentation tasks on two public datasets, and the results demonstrate the effectiveness of our method.

Keywords: Machine learning / Deep learning approaches, Ultrasound imaging - Prenatal, Image segmentation
Abstract: Good quality (annotated) data is one of the most important aspects of supervised deep learning. Tasks such as semantic segmentation have a huge data requirement in exchange for only satisfactory performance. Large-scale annotations spread across multiple annotators tends to create inconsistencies, as there are various manual and semi-automated techniques involved. This mandates an external evaluator or expert to check and narrow down the problematic annotations. Studies have shown that even marking a few instances wrong in classification can lead to a significant performance drop in the model (Mislabeling only 10% of one class can degrade the total performance of all classes by up to 10%). It has been noticed that fault localization by a medical expert is one of the most expensive and time-consuming processes. In this paper, we propose a novel framework for detecting the inconsistencies in the annotation of every object/anatomy in a specific image. We leverage the power of semi-supervised deep learning models (STCN) to help produce high-quality data for AI segmentation algorithms. Evaluation using this algorithm has been shown to reduce annotation review time by at least 5 hours for just 1000 images, and the quality of ground truth data improved thereby increasing the performance of the model by almost 3%.

Keywords: Machine learning / Deep learning approaches, Image classification, Ultrasound imaging - Cardiac
Abstract: Supervised deep learning has become defacto standard for most computer vision and machine learning problems including medical imaging. However, the requirement of having high quality annotations on large number of datasets places a huge overhead during model development. Self-supervised learning(SSL) is a paradigm which leverages unlabelled data to derive common-sense knowledge relying on signals present in the data itself for the learning rather than external supervisory signals. Recent times have seen the emergence of state-of-the-art SSL methods that have shown performance very close to supervised methods with minimal to no supervision on natural image settings. In this paper, we perform a thorough comparison of the performance of the state-of-the-art SSL methods for medical image setting, particularly for the challenging Cardiac view classification from Ultrasound acquisitions. We analyze the effect of data size in both phases of training - pre-text task training and main task training. We compare the performance with a task specific SSL technique based on simple image features and transfer learning ImageNet pre-training.

Keywords: Machine learning / Deep learning approaches, Image analysis and classification - Machine learning / Deep learning approaches, Image classification
Abstract: Dataset characteristics play an important role in training convolutional neural networks (CNNs) to evolve optimal features required to perform a specific task. Due to the high cost of recording and labelling surgical data, available datasets are relatively small in size and have been predominantly acquired at single sites. CNN-based approaches have been widely adapted to analyse surgical workflow using single-site datasets. Therefore, assessing generalised performance on data from different institutions has not been investigated. In this work, a CNN model that combines features from multiple stages to develop more accurate and generalised tool classification was introduced. An extensive evaluation of the proposed approach on three different datasets showed better generalised performance of our approach compared to base CNN models. The proposed approach achieved mAP values of 91.46%, 69.02% and 37.14% on the Cholec80, Cholec20 and Gyna05 datasets, respectively. The generalisation performance was improved on the achieved base CNN models mAP by about 7%.

Keywords: Machine learning / Deep learning approaches, Brain imaging and image analysis
Abstract: Traumatic brain injury (TBI) can drastically affect an individual's cognition, physical and emotional wellbeing, and behavior. Even patients with mild TBI (mTBI) may suffer from a variety of long-lasting symptoms, which motivates researchers to find better biomarkers. Machine learning algorithms have shown promising results in detecting mTBI from resting-state functional network connectivity (rsFNC) data. However, data collected at multiple sites introduces additional noise called siteeffects, resulting in erroneous conclusions. With the ongoing need to improve mTBI detection, this study shows that harmonization should be integrated into the machine learning process when working with multi-site neuroimaging datasets.

Keywords: Machine learning / Deep learning approaches, Image analysis and classification - Machine learning / Deep learning approaches, Multivariate image analysis
Abstract: In Radiomics, deep learning-based systems for medical image analysis play an increasing role. However, due to the better explainability, feature-based systems are still preferred, especially by physicians. Often, high-dimensional data and low sample size pose different challenges (e.g. increased risk of overfitting) to machine learning systems. By removing irrelevant and redundant features from the data, feature selection is an effective way of pre-processing. The research in this study is focused on unsupervised deep learning-based methods for feature selection. Five recently proposed algorithms are compared regarding their applicability and efficiency on seven data sets in three different sample applications. The exploration of distinctive features and the ability to rank their importance without the need for outcome information is a potential field of application for unsupervised feature selection methods. Especially in multiparametric radiology, the number of features is increasing. The identification of new potential biomarkers is important both for treatment and prevention. It was found that deep learning-based feature selection leads to improved classification results compared to conventional methods, especially for small feature subsets.

Keywords: Machine learning / Deep learning approaches, Magnetic resonance imaging - Cardiac imaging
Abstract: Abstract— Accurate quantification of myocardium strain in magnetic resonance images is important to correctly diagnose and monitor cardiac diseases. Currently, available methods to estimate motion are based on tracking brightness pattern differences between images. In cine-MR images, the myocardium interior presents an inhered homogeneity, which reduces the accuracy in estimated motion, and consequently strain. Neural networks have recently been shown to be an important tool for a variety of applications, including motion estimation. In this work, we investigate the feasibility of quantifying myocardium strain in cardiac resonance synthetic images using motion generated by a compact and powerful network called Pyramid, Warping, and Cost Volume (PWC). Using the motion generated by the neural network, the radial myocardium strain obtained presents a mean average error of 12.30% +- 6.50%, and in the circumferential direction 1.20% +- 0.61%, better than the two classical methods evaluated.

Keywords: Machine learning / Deep learning approaches, Image analysis and classification - Machine learning / Deep learning approaches, Image segmentation
Abstract: Fluorescence microscopy images of cell organelles enable the study of various complex biological processes. Recently, deep learning (DL) models are being used for the accurate automatic analysis of these images. DL models present state-of-the-art performance in many image analysis tasks such as object classification, segmentation and detection. However, to train a DL model a large manually annotated dataset is required. Manual annotation of 3D microscopy images is a time-consuming task and must be performed by specialists in the area. Thus, only a few images with annotations are typically available. Recent advances in generative adversarial networks (GANs) have allowed the translation of images with some conditions into realistic looking synthetic images. Therefore, in this work we explore approaches based on GANs to create synthetic 3D microscopy images. We compare four approaches that differ in the conditions of the input image. The quality of the generated images was assessed visually and using a quantitative objective GAN evaluation metric. The results showed that the GAN is able to generate synthetic images similar to the real ones. Hence, we have presented a method based on GANs to overcome the issue of small annotated datasets in the biomedical imaging field.

Keywords: Machine learning / Deep learning approaches, Image segmentation, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: In minimally invasive interventional surgery, ultrasound imaging is usually used to provide real-time feedback in order to obtain the best diagnostic results or realize treatment plans, so how to accurately obtain the position of the medical biopsy needle is a problem worthy of study. 2D ultrasound simulation images containing the medical biopsy needle are generated, and our images background is from the real breast ultrasound image. Based on the deep learning network, the images containing the medical biopsy needle are used to analyze the effectiveness of different networks for needle localization for the purpose of returning needle positions in non-uniform ultrasound images. The results show that attention U-Net performed best and can accurately reflect the real position of the medical biopsy needle. The IoU and Precision can reach 90.19% and 96.25%, and the Angular Error is 0.40°. Clinical Relevance— Based on the deep network, for 2D ultrasound images containing medical biopsy needle, the localization precision can reach 96.25% and the Angular Error is 0.40°.

Keywords: Machine learning / Deep learning approaches, Image segmentation, X-ray radiography
Abstract: This study aimed to determine a fundamental method for the automated detection and treatment of dental and orthodontic problems. Manual intervention is inefficient and prone to human error in detecting anomalies. Deep learning was used to identify a solution to this problem. We proposed leveraging incremental learning approaches using Mask R-CNN as backbone networks on small datasets to construct a more accurate model from automatically labeled data. The knowledge acquired at one stage of education is carried over to the subsequent stage. By incorporating newly annotated data, transfer learning improved the model's performance. Despite the data scarcity issues inherent in radiograph image collection, the findings for filling and tooth segmentation tasks were encouraging and adequate. We compared our results to prior research to optimize the performance of our proposed method.

Keywords: Deformable registration, Image registration, segmentation, compression and visualization - Volume rendering, Multimodal image fusion
Abstract: Image registration is a commonly required task in computer-assisted surgical procedures. Existing registration methods in laparoscopic navigation systems suffer from several constraints, such as a lack of deformation compensation. The proposed algorithm aims to provide the surgeons with updated navigational information about the deep-seated anatomy, which considers the continuous deformations in the operating environment. We extended an initial rigid registration to a shape-preserving deformable registration pathway by incorporating user interaction and an iterative mesh editing scheme that preserves local details. The proposed deformable registration workflow was tested with phantom and animal trial datasets. A qualitative evaluation based on expert feedback demonstrated a satisfactory outcome, and a commensurate execution efficiency was achieved. The improvements offered by the method, couples with its relatively easy implementation, make it an attractive method for adoption in future pre-clinical and clinical applications of augmented reality assisted surgeries.

Keywords: Image registration, segmentation, compression and visualization - Volume rendering, Rigid-body image registration
Abstract: Abstract— Augmented Reality (AR) can avoid some of the drawbacks of Minimally Invasive Surgery and may provide opportunities for developing innovative tools to assist surgeons. In laparoscopic surgery, the achievement of an easy and sufficiently accurate registration is an open challenge. This is particularly true in procedures, such as laparoscopic abdominal Sacro-Colpopexy, where there is a lack of a sufficient number of visible anatomical landmarks to be used as a reference for registration. In an attempt to address the above limitations, we present and preliminarily test a constrained manual procedure based on the identification of a single anatomical landmark in the laparoscopic images, and the intraoperative measurement of the laparoscope orientation. Tests in a rigid in-vitro environment show good accuracy (median error 2.4 mm obtained in about 4 min) and good preliminary feedback from the technical staff who tested the system. Further experimentation in a more realistic environment is needed to validate these positive results.

Clinical Relevance—This paper provides a new registration method for the development of AR educational videos and AR-based navigation systems for laparoscopic interventions.

Keywords: EEG imaging, Image reconstruction and enhancement - Machine learning / Deep learning approaches, Regularized image Reconstruction
Abstract: Brain source localization from electroencephalogram (EEG) signals is an challenging problem for noninvasively localizing the brain activity. Conventional methods use handcrafted regularization terms based on neural-physiological assumptions by exploiting the spatial-temporal structure on the source signals. In recent years, deep learning frameworks have demonstrated superior performance for solving the inverse problems in the natural and medical imaging field. This study proposes a novel unsupervised learning training-free framework based on Generative Adversarial Networks and deep image prior (GANs-DIP) as a generative model simulating spatially structured source signal. The proposed framework can faithfully recover extended source patches activation patterns of the brain in an unsupervised manner. Numerical experiments on a realistic brain model are performed under different levels of signal-to-noise ratio (SNR). The proposed model shows satisfactory performance in recovering the underlying source activation.

Keywords: Electrical impedance imaging, Functional image analysis, Ultrasound imaging - Elastography
Abstract: Here, we tested the feasibility of predicting CAP with multi-spectral EIT. Conductivity and CAP were acquired from nonalcoholic fatty liver disease patients using a portable EIT system and vibration-controlled transient elastography (VCTE). We then used frequency-difference conductivity and waist-over-height as prediction features to estimate CAP and found an adj. R2 of 0.92. We further developed a novel prediction method by incorporating EIT spectral unmixing reconstruction and demonstrated an improvement in CAP estimation. Last, we optimized the EIT acquisition process by minimizing the total variance of the CAP estimator.

Keywords: Electrical impedance imaging, Image reconstruction - Performance evaluation, Regularized image Reconstruction
Abstract: Incorporated with a structural prior, discrete cosine transformation (DCT) based electrical impedance tomography (EIT) algorithm can improve the interpretability of EIT images in clinical settings. However, this benefit comes with a risk of the untrue prior which yields a misleading result compromising clinical decision. The redistribution index is able to detect an untrue prior by analysing EIT reconstructions. In addition to structural priors, EIT reconstruction is also affected by the choice of hyperparameter λ in DCT-based EIT algorithm. In this research, influence of hyperparameter on untrue prior detection is investigated in terms of simulation experiment. A series of simulation settings consisting of 30 different atelectasis scales was conducted, then reconstructed with 20 different hyperparameters, to investigate the behavior of redistribution index. The result shows, despite the fact that redistribution index is indeed influenced by the choice of the hyperparameter λ, the detection of an untrue prior is not significantly affected. The untrue prior detection is rather stable regardless of the optimal hyperparameter. Clinical relevance—Optimal hyperparameter is not always guaranteed in clinical settings. This research confirms that the untrue prior detection is not strongly influenced by the hyperparameter. An update of untrue priors incorporated into EIT approach will facilitate a better interpretation of EIT results and an accurate clinical decision.

Keywords: Electrical impedance imaging
Abstract: Magnetic induction tomography (MIT) is harmless and contactless technique for measuring the conductivity of the biological tissue. MIT could be used for initial diagnosis and continuous monitoring of stroke. Different kinds of coil arrays have been proposed for MIT systems. Previous research results using a circular 16-channel MIT model reported difficulties with detection and measurement of small bioelectric signals. For stroke imaging, a system with a higher sensitivity is required. We aim to improve the sensitivity by increasing the number of coils and placing them closer to the head. In this paper, a helmet type coil array with 31 coils is introduced. For simplicity, the head is modelled as a sphere with white matter as a material. The stroke is simulated as a single inclusion with blood and assigned different sizes and positions. Sensitivity distribution and target response of the stroke were evaluated for the helmet model and compared with the circular MIT system. The simulations and analysis were performed at 10 MHz frequency with different coil pairs. Results from comparison of the two MIT models show that the Helmet coil array provides better spatial sensitivity, which has been estimated to be more than 20 times higher than the circular model. Further, when all coils are taken in account, the recorded sensitivity improvement was in the range of 13–90-fold.

Keywords: Image analysis and classification - Digital Pathology, Machine learning / Deep learning approaches, Optical imaging and microscopy - Super-resolution imaging
Abstract: Manual assessment of fragments during the processing of pathology specimens is critical to ensure that the material available for slide analysis matches that captured during grossing without losing valuable material during this process. However, this step is still performed manually, resulting in lost time and delays in making the complete case available for evaluation by the pathologist. To overcome this limitation, we developed an autonomous system that can detect and count the number of fragments contained on each slide. We applied and compared two different methods: conventional machine learning methods and deep convolutional network methods. For conventional machine learning methods, we tested a two-stage approach with a supervised classifier followed by unsupervised hierarchical clustering. In addition, Fast R-CNN and YOLOv5, two state-of-the-art deep learning models for detection, were used and compared. All experiments were performed on a dataset comprising 1276 images of colorectal biopsy and polypectomy specimens manually labeled for fragment/set detection. The best results were obtained with the YOLOv5 architecture with a map@0.5 of 0.977 for fragment/set detection.

Keywords: Image analysis and classification - Digital Pathology, Image analysis and classification - Machine learning / Deep learning approaches, Image classification
Abstract: Cervical cancer has become one of the important factors threatening women's health. Histopathological diagnosis is the most important criterion for cervical cancer diagnosis and treatment. Accurate classification of lesion degree of cervical epithelium by analyzing whole slide images (WSIs) can effectively improve the therapeutic effect and prognosis. However, classification of cervical lesion degree shows poor reproductivity due to lack of standardisation and is subjective among clinicians. In addition, due to the lack of large-scale finely annotated datasets, current deep learning methods do not perform well on this task. In this paper, we propose a two-stage method based on unsupervised pre-training to solve this multi-classification task. Our method first applied a patch-level network to predict the patch-level score and generate a heatmap that can highlight the lesion area. This network is pre-trained using an unsupervised method and verified on a public dataset. Then without extracting manual features, heatmaps are fed into a convolutional neural network (CNN) model directly for the WSI-level prediction. Our approach achieved an accuracy of 81.19% and a custom metric score of 0.9495 on the public cervical cancer WSI dataset, which is the highest in the public so far.

Keywords: Magnetic resonance imaging - Parallel MRI, Image enhancement - Denoising, Image reconstruction and enhancement - Machine learning / Deep learning approaches
Abstract: KerNL is a general kernel-based framework for autocalibrated reconstruction method, which does not need any explicit formulas of the kernel function for characterizing nonlinear relationships between acquired and unacquired k-space data. It is non-iterative without requiring a large amount of computational costs. Since the limited autocalibration signals (ACS) are acquired to perform KerNL calibration and the calibration suffers from the overfitting problem, more training data can improve the kernel model accuracy. In this work, virtual conjugate coil data are incorporated into the KerNL calibration and estimation process for enhancing reconstruction performance. Experimental results show that the proposed method can further suppress noise and aliasing artifacts with fewer ACS data and higher acceleration factors. Computation efficiency is still retained to keep fast reconstruction with the random projection.

Keywords: High throughput data - Machine learning and deep learning, High throughput data - Neural networks, support vector machine, and generative model
Abstract: Automatizing cranial implant design has become an increasingly important avenue in biomedical research. Benefits in terms of financial resources, time and patient safety necessitate the formulation of an efficient and accurate procedure for the same. This paper attempts to provide a new research direction to this problem, through an adversarial deep learning solution. Specifically, in this work, we present CranGAN - a 3D Conditional Generative Adversarial Network designed to reconstruct a 3D representation of a complete skull given its defective counterpart. A novel solution of employing point cloud representations instead of conventional 3D meshes and voxel grids is proposed. We provide both qualitative and quantitative analysis of our experiments with three separate GAN objectives, and compare the utility of two 3D reconstruction loss functions viz. Hausdorff Distance and Chamfer Distance. We hope that our work inspires further research in this direction.

Keywords: Model building - Algorithms and techniques for systems modeling
Abstract: Medical augmented reality and simulated test environments struggle in accurately simulating local sensor measurements across large spatial domains while maintaining the proper resolution of information required and real time capability. Here, a simple method for real-time simulation of intraoperative sensors is presented to aid with medical sensor development and professional training. During a surgical intervention, the interaction between medical sensor systems and tissue leads to mechanical deformation of the tissue. Through the inclusion of detailed finite element simulations in a real-time augmented reality system the method presented will allow for more accurate simulation of intraoperative sensor measurements that are independent of the mechanical state of the tissue. This concept uses a coarse, macro-level deformation mesh to maintain both computational speed and the illusion of reality and a simple geometric point mapping method to include detailed fine mesh information. The resulting system allows for flexible simulation of different types of localized sensor measurement techniques. Preliminary simulation results are provided using a real-time capable simulation environment and prove the feasibility of the method.

Keywords: Model building - Parameter estimation, Computational modeling - Analysis of high-throughput systems biology data, Model building - Sensitivity analysis
Abstract: Abstract— Metallic nanostructured-based biosensors provide label-free, multiplexed, and real-time detections of chemical and biological targets. Aluminum-based biosensors are favored in this category, due to their enhanced stability and profitability. Despite the recent advances in nanotechnology and the significant improvement in development of these biosensors, some deficiencies restrict their utilization. Hence a detailed insight into their behavior in different conditions would be crucial, which can be achieved with nanoscale numerical simulation. With this aim, an Aluminum-based biosensor is chosen to be analyzed with the help of all-atom molecular dynamics model (AA-MD), using large-scale atomic/molecular massively parallel simulator (LAMMPS). The surface properties and adsorption process through different flow conditions and various concentration of the target, are investigated in this study. In the future work, the results of this study will be used for developing a predictive model for surface properties of the biosensor.

Clinical Relevance— The role of biosensors in clinical applications and early diagnosis is evident. This work provides a model for predicting the binding behavior of the target molecules on the biosensor surface in different conditions. Results demonstrate an increase in the adsorption of ethanol on the biosensor surface of 7% up to 80% with changing the velocity from 0.001 m/s to 1 m/s Although for cases with higher concentration, this trend becomes complicated, necessitating the implementation of machine learning models in the future works.

Keywords: Models of medical devices, Model building - Parameter estimation, Model building - Algorithms and techniques for systems modeling
Abstract: The cardiopulmonary bypass system used in cardiac surgery can generate microbubbles (MBs) that may cause complications, such as neurocognitive dysfunction, when delivered into the blood vessel. Estimating the number of MBs generated, thus, is necessary to enable the surgeons to deal with it. To this end, we previously proposed a neural network-based model for estimating the number of MBs from four factors measurable from the cardiopulmonary bypass system: suction flow rate, venous reservoir level, blood viscosity, and perfusion flow rate. However, the model has not been adapted to the data collected from actual surgery cases. In this study, the accuracy of MBs estimated by the proposed model was examined in four clinical cases. The results showed that the coefficient of determination between estimated MBs and the measured MBs throughout the surgeries was R2=0.558 (p<0.001). We found that the surgical treatments, such as administration of drugs, fluids and blood transfusions, increased the number of measured MBs. The coefficient of determination increased to R2 = 0.8762 (p<0.001) by excluding the duration of these treatments. This result indicates that the model can estimate the number of MBs with high accuracy under the clinical environment.

Keywords: Models of medical devices, Organ modeling, Organs and medical devices - Multiscale modeling and the physiome
Abstract: Atherosclerosis is one of the most mortal diseases that affects the arterial vessels, due to accumulation of plaque, altering the hemodynamic environment of the artery by preventing the sufficient delivery of blood to other organs. Stents are expandable tubular wires, used as a treatment option. In silico studies have been extensively exploited towards examining the performance of such devices by employing Finite Element Modeling. This study models the crimping stage during stent implantation to examine the effect of inclusion of pre-stress state of the stent. The results show that modeling of the crimping stress state of the stent prior to the deployment results in under-expansion of the stent, due to the indirect inclusion of strain-induced hardening effects. As a result, it is evident that the compressive stent stress configuration is important to be considered in the computational modeling approaches of stent deployment.

Keywords: Models of medical devices, Organ modeling, Systems modeling - Decision making
Abstract: In this contribution, the development and validation of a digital twin of a decentralized source of breathing gas used for mechanical ventilation is described. A hardware setup is built in analogy to a Simscape ® model with almost identical architecture and geometry. An experimental trial of volume controlled ventilation was conducted and the obtained data of resulting pressure and oxygen concentration was compared. Correlating measured and simulated data lead to coefficients of 0.83 and 0.74 for FiO2 in the dosage chamber and airway, respectively. Clinical relevance— This work showed that the process of mechanical ventilation can be simulated even to the extent of oxygen distribution for ex-vivo applications.

Keywords: Organ modeling, Modeling of cell, tissue, and regenerative medicine - Tissue profiling , Synthetic biology
Abstract: Various studies have looked at the efficiency of artificial vessel and tissue networks in the study of photoplethysmography (PPG) in an effort to better understand the origin of various morphological features present in the signal. Whilst there are all reasonable attempts made to replicate geometrical features such as vessel depth, vessel wall thickness and diameter etc., not many studies have attempted to replicate the mechanical properties such as vessel elasticity and tissue compressibility. This study reports two methods for tissue mechanical testing for the analysis of vessel elasticity and tissue compressibility. A two-part polydimethylsiloxane (PDMS) was used as a base material for both tissue and vessel construction, and the properties altered by changing the curing component ratio. Tissue compression properties were investigated using an industrially calibrated materials testing device using the protocol from the ASTM 0575-91 testing method. Vessel elasticity was investigated using a custom method and apparatus to report vessel diameter and length change simultaneously. Tissue compressive properties proved reasonably easy to replicate through catalyst alteration, however the vessel elasticity properties were found to be higher than expected at all reasonable catalyst ratios. The property of hyper-elasticity was observed in the artificial vessels though, leading to the conclusion that alternative material recipes or construction methods may be needed to correctly replicate the expected mechanical characteristics.

Keywords: Organ modeling, Organs and medical devices - Multiscale modeling and the physiome
Abstract: Bariatric surgery presents a specific challenge in surgical education; simulators need to take into account the specific technical difficulties related to the patient population but also to various types of surgery. We interviewed several leaders in the fields of bariatric and general surgery with experience in laparoscopic surgery and developed a bariatric-specific laparoscopic simulator. This novel simulator was constructed using a variety of silicone-based materials and 3D printing techniques to be reusable and adjustable for a variety of procedures, with no essential components being disposed of following each use. Expert surgeons (n=3) with knowledge on bariatric procedures were recruited and asked to perform a simple simulated laparoscopic procedure. Following testing, participants were asked to complete a survey and rate the simulator based on its physical attributes, global realism, usefulness in improving surgical skills, and overall experience. Face and content validation outcomes based on the questionnaire evaluations completed by expert surgeons showed very good results, with an overall mean score of 4.2 out of 5 (84%). These preliminary results highlight the potential for the simulator’s application as a tool to improve bariatric surgical education and patient outcomes.

Keywords: Systems modeling - Patient stratification, Translational biomedical informatics - Decision making, Models of medical devices
Abstract: Abstract— Fetal surgery is regarded as a technically difficult and new field of research, requiring the use of fetoscopic and ultrasound (US) navigation to perform minimally invasive procedures within the amniotic cavity. The Surgical Apprenticeship Training model (SAT) centres around the subjective assessment of a surgical resident’s cognitive competency and technical skills under proctorship using opportunity-based environments. The restrictiveness and rarity of fetal procedures limit the effectiveness of the SAT model, resulting in a slow learning curve (LC) and higher procedural complication rates. This paper aimed to investigate the use of optical tracking technology to construct a novel simulated training system and accompanying scoring assessment under the Proficiency-Based Training model (PBT), providing real-time positional feedback of surgical tools and a quantitative feedback assessment of a surgical resident’s technical skills.

Clinical Relevance— Clinical feedback deemed the system as valid and confirmed that this novel approach to surgical training will significantly benefit smaller clinics that lack opportunity-based environments. Clinical feedback also suggested that the training system could be adapted to provide access to complex surgical training across the world.

Keywords: Models of organ physiology, Models of medical devices
Abstract: The interest in optical healthcare technologies has increased significantly over the recent years. The innovation of new optical technologies such as Near Infrared Spectroscopy (NIRS), used for the monitoring of brain perfusion, demands a comprehensive understanding and knowledge of the light tissue interaction. Phantoms can provide a rigorous, reproducible and convenient approach for evaluating an optical sensor’s performance. However, UpToDate literature does not provide a detailed description of a complete head model that involves the human anatomy, physiological changes, and the tissue optical properties. The later is key for the design, development and testing of optical sensors, such as NIRS technologies. This paper compared the optical properties of the materials chosen to build a head phantom, against the optical properties of real brain and skull tissues extracted from animal models. The spectra of a silicone brain and resin skull samples were compared with the spectra of the respective tissues extracted from pigs and mice. The results of this study demonstrated that both phantom materials have similar optical properties to mice and pigs’ tissues. The morphology of the phantom’s spectra were very similar to the respective animal tissue comparator.

Keywords: Cardiovascular, respiratory, and sleep devices - Implantables, Cardiovascular and respiratory signal processing - Cardiovascular signal processing, Cardiac mechanics, structure & function - Atrial Fibrillation
Abstract: In the context of increase in cardiovascular diseases in the aging population, including a high prevalence of atrial fibrillation (AF), the development of medical devices to ensure patient follow-up is of major interest. The purpose of this study is to assess the ECG signal quality of a one-lead in a new miniaturized device on healthy volunteers submitted to several conditions reflecting daily life activity. Our results show that the P wave identification is not enough reliable to consider the detection of its potential disappearance in case of AF. However, we show that the ECG signals can be used to robustly detect the RR intervals. To conclude, for rhythm disturbance detection, an automatic and specific analysis of RR variability has now to be integrated in this new multimodal device.

Keywords: Cardiovascular and respiratory signal processing - Cardiovascular signal processing, Cardiovascular regulation - Autonomic nervous system, Cardiovascular regulation - Heart rate variability
Abstract: Pulse rate variability (PRV) has been proposed as a surrogate for the estimation of Heart Rate Variability (HRV), which is a non-invasive technique used to assess the cardiac autonomic activity. However, both physiological and technical factors may affect the relationship between HRV and PRV, and there are no standards for the analysis of PRV from photoplethysmographic (PPG) signals. The aim of this study was to determine the best outlier management strategies for PRV analysis. 117 PPG signals with randomly generated PRV information were simulated using Gaussian signals. From these, interbeat intervals were detected and different outlier detection and correction techniques were applied. Time and frequency-domain and non-linear PRV indices were extracted and compared with respect to the gold standard values obtained from the simulated PRV information. The results show that, in good quality PPG signals, there is no need to apply any outlier management technique for the extraction of PRV information. Clinical relevance — Establishing guidelines for PRV measurement can lead to more reliable and comparable results, as well as to the increase in the use of this variable for the diagnosis and monitoring of cardiovascular and autonomic conditions.

Keywords: Cardiovascular and respiratory signal processing - Heart Rate and Blood Pressure Variability, Cardiovascular regulation - Heart rate variability, Cardiac mechanics, structure & function - Cardiac muscle mechanics
Abstract: Heart rate variability (HRV) is a physiological phenomenon of the variation of a cardiac interval (interbeat) over time that reflects the activity of the autonomic nervous system. HRV analysis is usually based on electrocardiograms (ECG signals) and has found many applications in the diagnosis of cardiac diseases, including valvular diseases. This analysis could also be performed on seismocardiograms (SCG signals) and gyrocardiograms (GCG signals) that provide information on cardiac cycles and the state of heart valves. In our study, we sought to evaluate the influence of valvular heart disease on the correlations between HRV indices obtained from electrocardiograms, seismocardiograms, and gyrocardiograms and to compare the HRV indices obtained from the three aforementioned cardiac signals. The results of HRV analysis in the time domain and frequency domain of the ECG, SCG, and GCG signals are within the standard deviation and have a strong linear correlation. This means that despite the influence of VHDs on the SCG and GCG waveforms, the HRV indices are valid. Clinical Relevance—Cardiac mechanical signals (seismocardiograms and gyrocardiograms) can be applied to evaluate heart rate variability despite the influence of valvular diseases on the morphology of cardiac mechanical signals.

Keywords: Cardiac electrophysiology - Ventricular arrhythmia mechanisms, Cellular and molecular cardiorespiratory engineering - Cellular arrhythmia mechanisms, Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: Abstract—Cardiac alternans has been associated with an increased propensity to lethal tachyarrhythmias such as ventricular tachycardia and fibrillation (VT/VF). Myocardial infarction (MI), resulting from restricted oxygen supply to the heart, is a known substrate for VT/VF. Here, we investigate the utility of cardiac alternans as a predictor of tachyarrhythmias in a chronic MI ovine model. In-vivo electrophysiological studies were performed to assess the change in microvolt T-wave alternans (TWA) with induction of acute ischemia following coronary artery occlusion. 24-hour telemetry was performed in an ambulatory animal for 6 weeks to monitor the progression of TWA with chronic MI. At 6 weeks, ex-vivo optical mapping experiments were performed to assess the spatiotemporal evolution of alternans in sham (n=5) and chronic MI hearts (n=8). Our results demonstrate that chronic MI leads to significant electrophysiological changes in the cardiac substrate. Significant increase in TWA is observed post occlusion and a steady rise in alternans is seen with progression of chronic MI. Compared to sham, chronic MI hearts show significant presence of localized action potential amplitude alternans, which spatially evolve with an increase in pacing frequency. Clinical Relevance—Our results demonstrate that localized alternans underlie arrhythmogenesis in chronic MI hearts and microvolt TWA can serve as a biomarker of disease progression during chronic MI.

Keywords: Cardiovascular and respiratory signal processing - Cardiovascular signal processing, Cardiac mechanics, structure & function - Cardiac muscle mechanics
Abstract: Heartbeat detection is an essential part of cardiac signal analysis because it is recognized as a representative measure of cardiac function. The gold standard for heartbeat detection is to locate QRS complexes in electrocardiograms. Due to the development of sensors and information and communication technologies (ICT), seismocardiography (SCG) is becoming a viable alternative to electrocardiography to monitor heart rate. In this work, we propose a system for detecting the heartbeat based on seismocardiograms using deep learning methods. The study was carried out with a publicly available data set (CEBS) that contains simultaneous measurements of ECG, breathing signal, and seismocardiograms. Our approach to heartbeat detection in seismocardiograms uses a model based on a ResNet-based convolutional neural network and contains a squeeze and excitation unit. Our model scored state-of-the-art results (Jaccard and F1 score above 97%) on the test dataset, demonstrating its high reliability.

Keywords: Sleep - Obstructive sleep apnea, Cardiovascular, respiratory, and sleep devices - Wearables, Sleep - Snoring
Abstract: Although sleep apnea is one of the most prevalent sleep disorders, most patients remain undiagnosed and untreated. The gold standard for sleep apnea diagnosis, polysomnography, has important limitations such as its high cost and complexity. This leads to a growing need for novel cost-effective systems. Mobile health tools and deep learning algorithms are nowadays being proposed as innovative solutions for automatic apnea detection. In this work, a convolutional neural network (CNN) is trained for the identification of apnea events from the spectrograms of audio signals recorded with a smartphone. A systematic comparison of the effect of different window sizes on the model performance is provided. According to the results, the best models are obtained with 60 s windows (sensitivity=0.72, specificity=0.89, AUROC=0.88). For smaller windows, the model performance can be negatively impacted, because the windows become shorter than most apnea events, by which sound reductions can no longer be appreciated. On the other hand, longer windows tend to include multiple or mixed events, that will confound the model. This careful trade-off demonstrates the importance of selecting a proper window size to obtain models with adequate predictive power. This paper shows that CNNs applied to smartphone audio signals can facilitate sleep apnea detection in a realistic setting and is a first step towards an automated method to assist sleep technicians.

Keywords: Vascular mechanics and hemodynamics - Arterial pressure in cardiovascular disease, Vascular mechanics and hemodynamics - Vascular Hemodynamics, Vascular mechanics and hemodynamics - Pulse wave velocity
Abstract: Pulse pressure (PP) is defined as the difference between systolic blood pressure (SBP) and diastolic blood pressure (DBP). The metric PP is not unique, as numerous combinations of SBP and DBP yield the same value for PP. Therefore, we introduced the PP companion (PPC) which is calculated using the Pythagorean theorem. Only the combination of PP and PPC offers unique characterization. Interestingly, PPC was found to be associated with mean arterial pressure (MAP). Another mathematical construct frequently used in hemodynamic studies refers to the ratio of DBP and SBP, or DBP/SBP, denoted as Prat. As Prat and PP share the same companion (C), we investigated the association between PratC and MAP, as well as the connection between PP and Prat. Various patient cohorts were included: A) 52 heart failure patients (16 women), B) 88 patients (11 women) with acute cardiac syndromes, C) 257 patients (68 men) diagnosed with atherosclerosis or any of various types of autoimmune disease, and D) 106 hypertensives (51 men). Linear regression analysis resulted in the following correlations: A: R (PratC, MAP) = 0.94, R (PP, Prat) = -0.91 B: R (PratC, MAP) = 0.98, R (PP, Prat) = -0.85 C: R (PratC, MAP) = 0.97, R (PP, Prat) = -0.86 D: R (PratC, MAP) = 0.92, R (PP, Prat) = -0.82 We conclude that Prat carries no substantial incremental value beyond PP, while both Prat and PP are incomplete metrics, requiring simultaneous consideration of MAP.

Keywords: Cardiovascular and respiratory signal processing - Blood pressure measurement, Cardiovascular regulation - Blood pressure variability
Abstract: Pulse arrival time (PAT), evaluated from electrocardiogram (ECG) and photoplethysmogram (PPG) signals, has been widely used for cuff-less blood pressure (BP) estimation due to its high correlation with BP. However, the question of whether filtering the PPG signal impacts the extracted PAT values and consequently, the correlation between PAT and BP, has not been investigated before. In this paper, using data from 18 subjects, changes in the PAT values, and in the subject-specific PAT-systolic BP (SBP) correlation caused by filtering the PPG signal with variable cutoff frequencies in the range of 2 to 15 Hz are studied. For PAT extraction, three PPG characteristic points (foot, maximum slope and systolic peak) are considered. Results show that differences in the cutoff frequency can shift the PAT values and introduce a worst-case error of over 8.2 mmHg for SBP estimation, indicating that PPG signal filter settings can impact PAT-based BP estimations. Our study suggests that extracting the PAT from the maximum slope point of PPG signal filtered at 10 Hz provides the most stable correlation with SBP.

Keywords: Cardiovascular regulation - Heart rate variability, Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: Premature infants are at risk of developing serious complications after birth. Communicative interventions performed in neonatal intensive care units (NICUs), such as music therapy interventions, can reduce the stress experienced by these infants and promote the development of their autonomic nervous system. In this study we investigated the effects of music therapy interventions, consisting of singing, humming, talking or rhythmic reading, on premature infants by investigating the effects on their heart rate variability (HRV). A total of 27 communicative intervention from 18 patients were included in this study. The NN-intervals were extracted from the ECG and the mean ± SEM values for the 6 different features (HR, SDNN, RMSSD, pNN50, pDec and SDDec) was investigated. Median feature values for the pre- and communicative intervention were compared using the Wilcoxon signed-rank test. An increase in values for the SDNN, RMSSD and pNN50 was found in the 20 minutes preceding the communicative intervention, when caregiving activities were performed, and was followed by an immediate decrease at the start of the intervention. Features’ variability during the intervention appeared to be smaller than in the pre-communicative intervention, indicating improved autonomic regulation. This difference was, however, not statistically significant possibly due to different types of activities applied during the communicative intervention per patient.