Keywords: Assistive and cognitive robotics in rehabilitation, Assistive and cognitive robotics in aided living, Home robots
Abstract: Cognitive disability strongly reduces people’s autonomy in performing desired as well as daily activities. The use of Social Assistive Robots (SARs) for cognitive rehabilitation therapy for disabled people could be a valuable gateway for the residential facility of the future. In this work, we design and develop a SAR that can be used for cognitive therapy proposing music and game activities. The results confirm that participants were positively engaged during the proposed activities and satisfied by the robot, despite the low perception of its usability. Professional caregivers noticed and confirmed the high level of engagement and the positive acceptance of the robot within the session, suggesting future tasks for SAR

Keywords: Human machine interfaces and robotics applications, Wearable robotic systems - Orthotics and Exoskeletons, Hardware and control developments in rehabilitation robotics
Abstract: Neurological trauma, such as stroke, traumatic brain injury (TBI), spinal cord injury, and cerebral palsy can cause mild to severe upper limb impairments. Hand impairment makes it difficult for individuals to complete activities of daily living, especially bimanual tasks. A robotic hand orthosis or hand exoskeleton can be used to restore partial function of an intact but impaired hand. It is common for upper extremity prostheses and orthoses to use electromyography (EMG) sensing as a method for the user to control their device. However some individuals with an intact but impaired hand may struggle to use a myoelectrically controlled device due to potentially confounding muscle activity. This study was conducted to evaluate the application of conventional EMG control techniques as a robotic orthosis/exoskeleton user input method for individuals with mild to severe hand impairments. Nine impaired subjects and ten healthy subjects were asked to perform repeated contractions of muscles in their forearm and then onset analysis and feature classification were used to determine the accuracy of the employed EMG techniques. The average accuracy for contraction identification across employed EMG techniques was 95.4% ± 4.9 for the healthy subjects and 73.9% ± 13.1 for the impaired subjects with a range of 47.0% ± 19.1 -- 91.6% ± 8.5. These preliminary results suggest that the conventional EMG control technologies employed in this paper may be difficult for some impaired individuals to use due to their unreliable muscle control.

Keywords: Joint biomechanics, Modeling and simulation in musculoskeletal biomechanics, Dynamics in musculoskeletal biomechanics
Abstract: Tele-operational tasks often suffer from instability issues and limited reliability during unpredictable interactions. We propose a real-time control law reproducing the impedance and kinematic behaviour of a subject’s arm (shoulder and elbow) on a remote avatar in a 2-DoF task. The human arm impedance and kinematics are estimated respectively from EMG and M-IMU data and then mapped into the avatar arm through an impedance control. Contrary to literature methods, our portable tele-impedance controller relies only on wearable sensors and enables an easy use in unstructured environments. The good performance (R2 > 0.7) of the muscle model used to map on the robot the human stiffness of five healthy subjects indicates the possibility of applying the proposed algorithm for a tele-impedance control.

Keywords: Wearable robotic systems - Orthotics and Exoskeletons, Exoskeleton applications
Abstract: The advent of wearable tremor suppression devices (WTSDs) has provided a promising alternative approach for parkinsonian tremor management, especially for individuals whose tremors are not managed by conventional treatment options. Currently, research in WTSDs has shown successful results with a tremor suppression ratio of up to 99%; however, the user safety of WTSDs has not been properly considered, especially in the occurrence of unexpected events, such as faults and disturbances. In this study, a fault-tolerant control system was developed and integrated into the control system of a WTSD for the first time. The safety and tremor suppression performance of the proposed system under the influence of a measurement loss fault were tested and evaluated on 18 tremor motion datasets, specifically by quantifying the tremor power suppression ratio and the error when tracking voluntary motion. The experimental evaluation showed that the proposed system could remain functional and safe to use in the existence of the fault, with an average user motion tracking error of 1.5◦. It was also found that the proposed system achieved significantly improved performance in both metrics when compared to the system without a fault-tolerant controller.

Keywords: Wearable robotic systems - Orthotics and Exoskeletons, Design and development of robots for human-robot interaction, Mechanics of locomotion and balance
Abstract: Humans are intrinsically unstable in quiet stance from a rigid body system viewpoint; however, they maintain balance thanks to neuro-muscular sensory properties whilst still exhibiting postural sway characteristics. This work introduces a one-degree-of-freedom supernumerary tail for balance augmentation in the sagittal plane to negate anterior-posterior postural sway. Simulations showed that the tail could successfully balance a human with impaired ankle stiffness and neural control. Insights into tail design and control were made; namely, to minimise muscular load the tail must have a significant component in the direction of the muscle, mounting location of the tail is significant in maximising inertial properties for balance augmentation and that adaptive control of the tail will be best suited for different loads held by a wearer.

Keywords: Assistive and cognitive robotics in aided living, Design and development of robots for human-robot interaction, Robot-aided mobility - Wheelchairs, canes, crutches, and mobility tools
Abstract: Intelligent assistive systems can navigate blind people, but most of them could only give non-intuitive cues or inefficient guidance. Based on computer vision and vibrotactile encoding, this paper presents an interactive system that provides blind people with intuitive spatial cognition. Different from the traditional auditory feedback strategy based on speech cues, this paper firstly introduces a vibration-encoded feedback method that leverages the haptic neural pathway and enables the users to interact with objects other than manipulating an assistance device. Based on this strategy, a wearable visual module based on an RGB-D camera is adopted for 3D spatial object localization, which contributes to accurate perception and quick object localization in the real environment. The experimental results on target blind individuals indicate that vibrotactile feedback reduces the task completion time by over 25% compared with the mainstream voice prompt feedback scheme. The proposed object localization system provides a more intuitive spatial navigation and comfortable wearability for blindness assistance.

Keywords: Physiological monitoring - Instrumentation, Bio-electric sensors - Sensor systems, Wearable sensor systems - User centered design and applications
Abstract: Bioimpedance has emerged as a promising modality to continuously monitor hemodynamic and respiratory physiological parameters through a non-invasive skin-contact approach. Bioimpedance sensors placed at the radial zone of the volar wrist provide sensitive operation to the blood flow of the underlying radial artery. The translation of bioimpedance systems into medical-grade settings for continuous hemodynamic monitoring, however, presents challenges when constraining the necessary sensing components to a minimal form factor while maintaining sufficient accuracy and precision of measurements. Thus, it is important to understand the effects of electrode configuration on bioimpedance signals when reducing them to a wearable form factor. Previous work regarding electrode configurations in bioimpedance does not address wearable constraints, nor do they focus on electrodes viable for wearable applications. In this study, we present empirical evidence of the effects of dry silver electrode sizes and spacings on the specificity and sensitivity of a wrist-worn bioimpedance sensor array. We found that wrist-worn bioimpedance systems for hemodynamic monitoring would benefit from reduced injection electrode spacings (up to a 392% increase in signal amplitude with a 50% decrease in spacing), increased sensing electrode spacings, and decreased electrode surface areas.

Keywords: Wearable body sensor networks and telemetric systems, IoT sensors for health monitoring, Health monitoring applications
Abstract: Existing Electrocardiogram (ECG) systems are either wired or based on radiofrequency (RF) wireless devices when remote transmission is needed. However, the use of radiofrequencies has limitations especially for sensitive populations such as newborns and infants. In addition, due to electromagnetic interference problems, impairments in the RF transmission of the ECG signal can lead to diagnostic errors. To answer these problems, we propose in this article, an optical wireless monitoring system, using an infrared link between an ECG sensor placed on a baby's chest and receivers placed on the ceiling of a pediatric room. In addition, it is assumed that the infant can move around in his bed. Our main contribution is the evaluation of the quality of the ECG signal transmitted by the proposed system, in terms of classic Signal to Noise Ratio (SNR) and Bit Error Rate (BER) metrics but also in terms of Signal Quality Indexes (SQIs) calculated from the characteristics of the received ECG signal. Results show the ability to perform wireless infant ECG monitoring using the Optical Wireless Communication (OWC) with satisfactory quality for optimum power.

Keywords: Physiological monitoring - Novel methods, Physiological monitoring - Modeling and analysis, Mechanical sensors and systems
Abstract: Blood pressure (BP) is a key parameter in critical care and in cardiovascular disease management. BP is typically measured via cuff-based oscillometry. This method is highly inaccurate in hypo-and hypertensive patients. Improvements are difficult to achieve because oscillometry is not yet fully understood; many assumptions and uncertainties exist in models describing the process by which arterial pulsations become expressed within the cuff signal. As a result, it is also difficult to estimate other parameters via the cuff such as arterial stiffness, cardiac output and pulse wave velocity (PWV)-BP calibration. Many research modalities have been employed to study oscillometry (ultrasound, computer simulations, ex-vivo studies, measurement of PWV, mechanical analysis). However, uncertainties remain; additional investigation modalities are needed. In this study, we explore the extent to which MRI can help investigate oscillometric assumptions. Four healthy volunteers underwent a number of MRI scans of the upper arm during cuff inflation. It is found that MRI provides a novel perspective over oscillometry; the artery, surrounding tissue, veins and the cuff can be simultaneously observed along the entire length of the upper arm. Several existing assumptions are challenged: tissue compression is not isotropic, arterial transmural pressure is not uniform along the length of the cuff and propagation of arterial pulsations through tissue is likely impacted by patient-specific characteristics (vasculature position and tissue composition). Clinical Relevance— The cuff interaction with the vasculature is extremely complex; existing models are oversimplified. MRI is a valuable tool for further development of cuff-based physiological measurements.

Keywords: Wearable body sensor networks and telemetric systems, Physiological monitoring - Instrumentation, Health monitoring applications
Abstract: Wearable devices for continuous non-invasive blood pressure monitoring must be capable of providing a continuous waveform representative of arterial blood pressure. This paper establishes the distinctions in waveform morphology between wearable sensor modalities, specifically millimeter-wave radar and photoplethysmography, when compared to a reference continuous non-invasive blood pressure monitor. An analysis of a 115-subject dataset was conducted to assess waveform suitability. Millimeter-wave radar waveform morphology was found to more closely resemble continuous non-invasive blood pressure than photoplethysmography.

Keywords: Wearable low power, wireless sensing methods, New sensing techniques, Physiological monitoring - Novel methods
Abstract: Abstract— Early detection of cardiovascular diseases via non-invasive, convenient, and continuous monitoring is crucial to reducing preventable deaths. This paper illustrates such monitoring using wearable near-field radio-frequency sensors to analyze ventricle and valve transients, which can be used as indicators of myriad cardiac disorders. We applied a novel vector injection signal processing method to improve timing consistency in ventricular contraction, ventricular relaxation, and valve opening extraction. The median relative timing error in valve opening detection was 14.7ms and 37.8ms for semilunar and atrioventricular valves, respectively, as benchmarked by the S1 and S2 heart sounds from a synchronous phonocardiogram. Clinical Relevance— No wearable sensor currently exists to conveniently and reliably evaluate ventricular and valvular dynamics, specifically valvular opening. Beyond extraction of the heart rate and its variation, the method in this paper has the potential to enable non-invasive measurements of detailed cardiac cycle timing features including valve openings, isovolumetric contraction/relaxation times, and ejection periods, improving the monitoring of patient health away from clinical healthcare centers.

Keywords: Health monitoring applications, Physiological monitoring - Modeling and analysis, IoT sensors for health monitoring
Abstract: In dialysis patients, monitoring vascular flow of the surgically created arteriovenous fistula (AVF) is critical to indicate the success of the AVF as a dialysis access site. Current gold standard to quantify vascular flow involves external doppler evaluation which requires frequent visits to the clinic. In this paper, we present a proof-of-concept cost-efficient vascular flow monitoring system towards a wearable and robust blood flow monitoring system. The proposed system captures beat-to-beat blood flow from impedance plethysmography (IPG) signal and performs embedded computing to robustly map the changes in the IPG to peripheral blood flow. We present the proof-of-concept results for the embedded real-time blood flow computing from measurements obtained using a custom electrical bioimpedance hardware presented previously elsewhere. We anticipate the results serving as the first step towards potentially eliminating the need for using expensive and bulky systems that require specialized personnel to operate for peripheral blood flow monitoring.

Keywords: Computational modeling - Analysis of high-throughput systems biology data, High throughput data - Machine learning and deep learning, Systems biology and systems medicine - prediction of disease related regulator
Abstract: The identification of protein target and mechanism of disease are fundamentally important in drug discovery. A pipeline for predicting mechanism of action (MoA) for drug molecules based on gene expression and cell viability data is developed and demonstrated on experimental data. A deep learning network learns the known MoAs over thousands of gene expression and cell viability training data and is shown to predict the unknown MoAs over test data with high efficacy.

Keywords: Computational modeling - Analysis of high-throughput systems biology data, Systems biology and systems medicine - RNA-seq analysis
Abstract: Single-cell RNA sequencing is a powerful method that helps delineate the regulatory mechanisms shaping the diverse cellular populations. Heterogeneous cell populations consist of individual cells, and the expression of distinct sets of genes can differentiate one sub-population of cells from another, as they are responsible for the emergence of distinct cellular phenotypes. Of particular importance are cells at transition states that bridge these different cellular phenotypes. In this study, we develop a method to identify the cells in transition states bridging different cellular phenotypes. Our approach is based on persistent homology, which enabled us to identify the group of cells located on the boundaries between different sub-populations of cells. We applied this method to study the reprogramming of human fibroblasts toward induced pluripotent stem cells using single-cell time-course data. Even though only the data that is representative of the early stages of the reprogramming process are analyzed, we are able to uncover transient cells bridging different cell sub-populations. The most prominent group of transient cells are found to be enriched for NANOG which is a known stem cell transcription factor that takes part in the maintenance of pluripotency and other stem cell marker genes. Overall, our method can identify cells in transient states bridging major cellular phenotypes, even though they are only a small fraction of the overall cell population. We also discuss how this approach can link the topology of the surface of cellular transcripts and bring order to the transition between cellular states, and how it automatically uncovers the underlying time process.

Keywords: Systems modeling - Patient stratification, High throughput data - Machine learning and deep learning, High throughput data - Pattern recognition
Abstract: Identifying rare tuberous sclerosis complex (TSC) children is valuable and crucial. Magnetic resonance imaging (MRI) is used for rare TSC diagnoses. In this work, T2w and FLAIR were combined as a new modality named FLAIR3 to maximize the contrast between TSC lesions and normal-appearing brain tissues. After that, for the first time, we propose to use two different 3D CNN combined with late fusion strategies to diagnose TSC. A total of 520 children were enrolled in the study, including 260 health and 260 TSC children. The experiments had shown that the FLAIR3 could effectively improve the conspicuity of TSC lesions and classification performance. And the results showed the proposed late fusion method can improve the classification performance and achieve the state-of-the-art performance of the AUC of 0.994 and the accuracy of 0.971, which could be treated as an effective computer-aided diagnostic tool to help clinical radiologists diagnose TSC children.

Keywords: High throughput data - Machine learning and deep learning
Abstract: During consultation dermatologists have to address hundreds of lesions in a limited amount of time. They will not only evaluate the single lesion of interest but more importantly the context of it. Visually comparing the similarity of the majority of lesions within the same patient provides a strong indication for lesions with significantly differing aspects. Deep learning algorithms are capable to identify such outliers, i.e. images that differ considerably from the expected appearance on a larger cohort, and highlight the main differences in those cases. In the present study we evaluate the use of autoencoders as unsupervised tools to detect suspicious skin lesions based on evaluation of real world data acquired during consultation at the USZ Dermatology Clinic.

Keywords: Model building - Signal and pattern recognition, Computational modeling - Analysis of high-throughput systems biology data
Abstract: Seizure termination has received significantly less attention than initiation and propagation and consequently, remains a poorly understood phase of seizure evolution. Yet,its study may have a significant impact on the development of efficient interventional approaches, i.e., it may be critical for the design of treatments that induce or reproduce termination mechanisms that are triggered in self-terminating seizures. In this work, we aim to study temporal and spectral features of intracranial EEG (iEEG) during epileptic seizures to find time-frequency signatures that can predict the termination patterns. We propose a deep learning model for classification of multi channel iEEG epileptic seizure termination pattern into burst suppression and continuous bursting. We decompose the raw time series seizure data into time-frequency maps using Morlet Wavelet Transform. A Convolution Neural Network(CNN) is then trained on cross-patient time-frequency maps to classify the seizure termination patterns. For evaluation of classification performance, we compared the proposed method with k-Nearest Neighbour (k-NN). The CNN is shown to achieve an accuracy of 90% and precision of 92% as compared to 70% accuracy achieved with the k-NN. The proposed model is thus able to capture the temporal and spatial patterns which results in high performance of the classifier. This method of classification can be used to predict how a particular seizure will end and can potentially inform seizure management and treatment.

Keywords: Systems modeling - Clinical applications of biological networks, Computational modeling - Biological networks, Data-driven modeling
Abstract: Cognitive control, the ability to rapidly shift one's attention and behavioral strategy in response to environmental changes, is often compromised across psychiatric disorders. One of the well-validated behavioral paradigms for tapping into the cognitive control circuits is a cognitive interference task, where subjects must suppress a natural response to follow a less intuitive rule. Slower response times on these tasks indicate difficulty exerting control to overcome response conflict. Conflict evokes robust electrophysiological signatures, such as theta (4-8 Hz) oscillations in the prefrontal cortex (PFC). However, the underlying neural mechanisms of conflict-evoked theta oscillations in the PFC are not clear. The objective of this work is to use a neural mass model (NMM) to find feasible cortical networks generating theta oscillations during conflict processing in human subjects. We used intracranial EEG (iEEG) recorded from dorsolateral PFC (dlPFC) and lateral temporal lobe (LTL) of human subjects with intractable epilepsy undergoing invasive monitoring, while they performed a multi-source interference task (MSIT). We used a dynamic causal modeling (DCM) framework to simulate dlPFC-LTL theta using a Jansen-Rit NMM. We found significant evidence for an LTL input into the dlPFC during the initial 500 ms of conflict processing compared to a bidirectional connection between the dlPFC and LTL. We conclude that a neural mass modeling framework can be used to elucidate candidate mechanisms of neural oscillations underlying conflict resolution in human subjects.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification
Abstract: Rapid eye movement (REM) sleep behavior disorder (RBD) is parasomnia and a prodromal manifestation of Parkinson’s disease. The current diagnostic method relies on manual scoring of polysomnograms (PSGs), a procedure that is time and effort intensive, subject to interscorer variability, and requires high level of expertise. Here, we present an automatic and interpretable diagnostic tool for RBD that analyzes PSGs using end-to-end deep neural networks. We optimized hierarchical attention networks in a 5-fold cross validation directly to classify RBD from PSG data recorded in 143 participants with RBD and 147 age- and sex-matched controls. An ensemble model using logistic regression was implemented to fuse decisions from networks trained in various signal combinations. We interpreted the networks using gradient SHAP that attribute relevance of input signals to model decisions. The ensemble model achieved a sensitivity of 91.4 % and a specificity of 86.3 %. Interpretation showed that electroencephalography (EEG) and leg electromyography (EMG) exhibited most patterns with high relevance. This study validates a robust diagnostic tool for RBD and proposes an interpretable and fully automatic framework for end-to-end modeling of other sleep disorders from PSG data. Clinical relevance – This study presents a novel diagnostic tool for RBD that considers neurophysiologic biomarkers in multiple modalities.

Keywords: Data mining and big data methods - Pattern recognition, Neural networks and support vector machines in biosignal processing and classification
Abstract: Nowadays, high amounts of data can be acquired in various applications, spurring the need for interpretable data representations that provide actionable insights. Algorithms that yield such representations ideally require as little a priori knowledge about the data or corresponding annotations as possible. To this end, we here investigate the use of Kohonen’s Self-Organizing Map (SOM) in combination with data-driven low-dimensional embeddings obtained through self-supervised Contrastive Predictive Coding. We compare our approach to embeddings found with an auto-encoder and, moreover, investigate three ways to deal with node selection during SOM optimization. As a challenging experiment we analyze nocturnal sleep recordings of healthy subjects, and conclude that - for this noisy real-life data -contrastive learning yields a better low-dimensional embedding for the purpose of SOM training, compared to an auto-encoder. In addition, we show that a stochastic temperature-annealed SOM-training outperforms both a deterministic and a non-temperature-annealed stochastic approach.

Clinical relevance: The hypnogram has for decades been the clinical standard in sleep medicine, despite the fact that it is a highly simplified representation of a polysomnography recording. We propose a sensor-agnostic algorithm that is able to reveal more intricate patterns in sleep recordings, which might teach us about sleep structure and sleep disorders.

Keywords: Time-frequency and time-scale analysis - Wavelets, Physiological systems modeling - Multivariate signal processing
Abstract: Because drowsiness is a major cause in vehicle accidents, its automated detection is critical. Scale-free temporal dynamics is known to be typical of physiological and body rhythms. The present work quantifies the benefits of applying a recent and original multivariate selfsimilarity analysis to several modalities of polysomnographic measurements (heart rate, blood pressure, electroencephalogram and respiration), from the MIT-BIH Polysomnographic Database, to better classify drowsiness-related sleep stages. This study shows that probing jointly temporal dynamics amongst polysomnographic measurements, with a proposed original multivariate multiscale approach, yields a gain of above 5% in the Area-under-Curve quantifying drowsiness-related sleep stage classification performance compared to univariate analysis.

Keywords: Causality, Coupling and synchronization - Coherence in biomedical signal processing, Nonlinear dynamic analysis - Biomedical signals
Abstract: Information flow existed across brain regions, and varies dynamically during sleep. In evaluating brain communication and neural-oscillation connectivity across spatiotemporal scales, the phase-amplitude coupling (PAC) is well-explored. However, the directional connectivity is still a deficiency. In this work, we propose a cross-phase-amplitude transfer entropy method in quantifying the characteristics of multi-regional sleep dynamics. The simulation of multivariate nonlinear and nonstationary signals verifies both effectiveness and veracity of the proposed algorithm. The results achieved in sleep EEG of healthy adults indicate that the direction of PAC is from the occipital lobe to the frontal lobe in the Awake and N1 sleep stages. And the flow of PAC turns to the opposite direction for the other sleep stages, i.e., frontal-to-occipital lobe. Besides, the δ-θ/α PAC gradually strengthens with the deepening of the sleep. Of note, the PAC results in the REM sleep stage vary across different frequency pairs. The obtained results support the proposed method as a reliable tool in evaluating brain functions during sleep with brain signals.

Keywords: Physiological systems modeling - Signal processing in physiological systems, Multivariate methods, Signal pattern classification
Abstract: Previous studies have suggested that the typical slow oscillations (SO) characteristics during sleep could be modified in the presence of pediatric obstructive sleep apnea (OSA). Here, we evaluate whether these modifications are significant and if they may reflect cognitive deficits. We recorded the overnight electroencephalogram (EEG) of 294 pediatric subjects (5-9 years old) using eight channels. Then, we divided the cohort in three OSA severity groups (no OSA, mild, and moderate/severe) to characterize the corresponding SO using the spectral maximum in the slow wave sleep (SWS) band delta1 0.1-2 Hz (Max_SO), as well as the frequency where this maximum is located (FreqMax_SO). Spectral entropy (SpecEn) from delta1 was also included in the analyses. A correlation analysis was performed to evaluate associations of these spectral measures with six OSA-related clinical variables and six cognitive scores. Our results indicate that Max_SO could be used as a moderate/severe OSA biomarker while providing useful information regarding verbal and visuo-spatial impairments, and that FreqMax_SO emerges as an even more robust indicator of visuospatial function. In addition, we uncovered novel insights regarding the ability of SpecEn in delta1 to characterize OSA-related disruption of sleep homeostasis. Our findings suggest that the information from SO may be useful to automatically characterize moderate/severe pediatric OSA and its cognitive consequences.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Inter-subject variability and personalized approaches, Data mining and big data methods - Biosignal classification
Abstract: In this work we introduce a novel meta-learning method for sleep scoring based on self-supervised learning. Our approach aims at building models for sleep scoring that can generalize across different patients and recording facilities, but do not require a further adaptation step to the target data. Towards this goal, we build our method on top of the Model Agnostic Meta-Learning (MAML) framework by incorporating a self-supervised learning (SSL) stage, and call it S2MAML. In our analysis, we show that S2MAML can significantly outperform MAML. The gain in performance comes from the SSL stage, which we base on a general purpose pseudo-task that limits the overfitting to the subject-specific patterns present in the training dataset. We show that S2MAML outperforms standard supervised learning and MAML on the SC, ST, ISRUC, UCD and CAP datasets.

Keywords: Sensor Informatics - Physiological monitoring, Sensor Informatics - Sensor-based mHealth applications, Imaging Informatics - Image registration, segmentation, and compression
Abstract: In-vehicle health monitoring allows for continuous vital sign measurement in everyday life. Eventually, this could lead to early detection of cardiovascular diseases. In this work, we propose non-contact heart rate (HR) monitoring utilizing near-infrared (NIR) camera technology. Ten healthy volunteers are monitored in a realistic driving simulator during resting (5 min) and driving (10 min). We synchronously acquire videos using an out-of-the-shelf, low-cost NIR camera and 3-lead electrocardiography (ECG) serves as ground truth. The MediaPipe face detector delivers the region of interest (ROI) and we determine the HR from the peak with maximum amplitude within the power spectrum of skin color changes. We compare video-based with ECG-based HR, resulting in a mean absolute error (MAE) of 7.8 bpm and 13.0 bpm in resting and driving condition, respectively. As we apply only a simple signal processing pipeline without sophisticated filtering, we conclude that NIR camera-based HR measurements enables unobtrusive and non-contact monitoring to a certain extent, but artifacts from subject movement pose a challenge. If these issues can be addressed, continuous vital sign measurement in everyday life could become reality.

Keywords: Sensor Informatics - Sensors and sensor systems, General and theoretical informatics - Data storage, Sensor Informatics - Wireless sensors and systems
Abstract: With the enormous amount of data collected by unobtrusive sensors, the potential of utilizing these data and applying various multi-modal advanced analytics on them is numerous and promising. However, taking advantage of the ever-growing data requires high-performance data-handling systems to enable high data scalability and easy data accessibility. This paper demonstrates robust design, developments, and techniques of a hierarchical time-indexed database for decision support systems leveraging irregular and sporadic time series data from sensor systems, e.g., wearables or environmental. We propose a technique that leverages the flexibility of general purpose, high-scalability database systems, while integrating data analytics focused column stores that leverage hierarchical time indexing, compression, and dense raw numeric data storage. We have evaluated the performance characteristics and tradeoffs of each to understand the data access latencies and storage requirements, which are key elements for capacity planning for scalable systems.

Keywords: Sensor Informatics - Wearable systems and sensors, Sensor Informatics - Physiological monitoring, Sensor Informatics - Sensor-based mHealth applications
Abstract: In modern psychotherapy, digital health technology offers advanced and personalized therapy options, increasing availability as well as ecological validity. These aspects have proven to be highly relevant for children and adolescents with obsessive-compulsive disorder (OCD). Exposure and Response Prevention therapy, which is the state-of-the-art treatment for OCD, builds on the reconstruction of everyday life exposure to anxious situations. However, while compulsive behavior predominantly occurs in home environments, exposure situations during therapy are limited to clinical settings. Telemedical treatment allows to shift from this limited exposure reconstruction to exposure situations in real life. In the SSTeP KiZ study (smart sensor technology in telepsychotherapy for children and adolescents with OCD), we combine video therapy with wearable sensors delivering physiological and behavioral measures to objectively determine the stress level of patients. The setup allows to gain information from exposure to stress in a realistic environment both during and outside of therapy sessions. In a first pilot study, we explored the sensitivity of individual sensor modalities to different levels of stress and anxiety. For this, we captured the obsessive-compulsive behavior of five adolescents with an ECG chest belt, inertial sensors capturing hand movements, and an eye tracker. Despite their prototypical nature, our results deliver strong evidence that the examined sensor modalities yield biomarkers allowing for personalized detection and quantification of stress and anxiety. This opens up future possibilities to evaluate the severity of individual compulsive behavior based on multi-variate state classification in real-life situations.

Keywords: Sensor Informatics - Wearable systems and sensors, Sensor Informatics - Multi-sensor data fusion, Sensor Informatics - Intelligent medical devices and sensors
Abstract: In this work, a novel multi-modal device that allows data to simultaneously be collected from three noninvasive sensor modalities was created. Force myography (FMG), surface electromyography (sEMG), and inertial measurement unit (IMU) sensors were integrated into a wearable armband and used to collect signal data while subjects performed gestures important for the activities of daily living (ADL). An established machine learning algorithm was used to decipher the signals to predict the user's intent/gesture being held, which could be used to control a prosthetic device. Using all three modalities provided statistically-significant improvements over most other modality combinations, as it provided the most accurate and consistent classification results.

Clinical relevance—The use of three sensing modalities can improve gesture-based control of upper-limb prosthetics.

Keywords: Sensor Informatics - Wearable systems and sensors, General and theoretical informatics - Machine learning
Abstract: Understanding how macronutrients (e.g., carbohydrates, protein, fat) affect blood glucose is of broad interest in health and dietary research. The general effects are well known, e.g., adding protein and fat to a carbohydrate-based meal tend to reduce blood glucose. However, there are large individual differences in food metabolism, to where the same meal can lead to different glucose responses across individuals. To address this problem, we present a technique that can be used to simultaneously (1) model macronutrients’ effects on glucose levels over time and (2) capture inter-individual differences in macronutrient metabolism. The technique performs a linear decomposition of glucose responses, alternating between estimating the macronutrients’ effect over time and capturing an individual’s sensitivity to macronutrients. On an experimental dataset containing glucose responses to a variety of mixed meals, the technique is able to extract basis functions for the macronutrients that are consistent with their hypothesized effects on PPGRs, and also characterize how macronutrients affect individuals differently.

Keywords: Sensor Informatics - Wearable systems and sensors, Sensor Informatics - Data inference, mining, and trend analysis, General and theoretical informatics - Pattern recognition
Abstract: The choice of appropriate machine learning algorithms is crucial for classification problems. This study compares the performance of state-of-the-art time-series deep learning algorithms for classifying food intake using sensor signals. The sensor signals were collected with the help of a wearable sensor system (the Automatic Ingestion Monitor v2, or AIM-2). AIM-2 used an optical and 3-axis accelerometer sensor to capture temporalis muscle activation. Raw signals from those sensors were used to train five classifiers (multilayer perceptron (MLP), time Convolutional Neural Network (time-CNN), Fully Convolutional Neural Network (FCN), Residual Neural Network (ResNet), and Inception network) to differentiate food intake (eating and drinking) from other activities. Data were collected from 17 pilot subjects over the course of 23 days in free-living conditions. A leave one subject out cross-validation scheme was used for training and testing. Time-CNN, FCN, ResNet, and Inception achieved average balanced classification accuracy of 88.84%, 90.18%, 93.47%, and 92.15%, respectively. The results indicate that ResNet outperforms other state-of-the-art deep learning algorithms for this specific problem.

Keywords: Robotic-aided therapies - Computer-assisted surgery systems, Image-guided therapies - Biopsy systems & technologies
Abstract: To improve the outcome of minimally invasive renal interventions, traditional video-guided needle navigation can be enhanced by tracking the needle, guiding the needle using video imaging, and augmenting the surgical scene with pre-procedural images or models of the anatomy. In our previous work we studied, both through simulations and in vitro experiments, the uncertainty associated with the model-to-phantom registration, as well as the camera-tracker calibration and video-guided navigation. In this work, we characterize the overall navigation uncertainty using tissue emulating patient-specific kidney phantoms featuring both virtual and physical internal targets. Pre-procedural models of the kidney phantoms and internal targets are generated from cone-beam CT images, and are registered to their intra-operative physical counterparts. The user then guides the needle insertion to reach the internal targets using video-based imaging augmented with a virtual representation of the needle tracked in real time. Following navigation, we acquire post-procedural cone-beam CT images of the phantoms and inserted needles. These images are used to determine the ground truth needle navigation accuracy (i.e., needle to target distance) against which the intra-operative navigation accuracy (i.e., intra-op needle tip to target distance) is assessed. We also explore a method to update the pre-procedural model to physical phantom registration intra-operatively using tracked video imaging, with the overall goal to improve overall navigation accuracy in the event of sub-optimal initial image-to-phantom registration. Our results showed a navigation error of less than 3.5 mm in gelatin phantoms and less than 6.5 mm in PVA phantoms. Following registration correction intra-operatively, we showed an overall improvement in navigation from roughly 6 mm RMS to approximately 2 mm RMS error, which is acceptable given the inherent tracking, 3D printing and phantom manufacturing limitations.

Keywords: FNIR (functional near infra-red) spectroscopy and near-infrared scanning and assessment, Ambulatory Diagnostic devices - Point of care technologies, Clinical engineering -Verification and validation of diagnostic & therapeutic systems / technologies
Abstract: This paper reports the application of a low-cost diagnostic modality for fat analysis in a liver phantom as well as human liver donors. The device works on the principle of diffuse reflectance spectroscopy, which absorbs and/or scatters depending upon the molecules that compose a tissue. Here, we describe the development of liver phantom of varying fat concentration using saturated fat mimicking liver steatosis. Followed by a pilot study in the human liver donor setting. Later, handheld device based on Infrared-LED and Photodetector for real-time time assessment of live donor liver and fat assessment.

Clinical Relevance— This device can be used in the development of an accurate and non-invasive for quantification of liver fat in the deceased donor selection process. It has an error margin of 10% in the quantification of fat which is comparable to a standard biopsy technique.

Keywords: Clinical engineering -Verification and validation of diagnostic & therapeutic systems / technologies, Defibrillators (implantable or external), Implantable devices for cardiac monitoring
Abstract: Implantable cardioverter defibrillators (ICDs) are developed to provide timely therapies when adverse patient conditions are detected. Device therapies need to be adjusted for individual patients and evolving patient conditions, which can be achieved by adjusting device parameter settings. However, there are no validated clinical guidelines for parameter personalization, especially for patients with complex and rare conditions. In this paper, we propose a reinforcement learning framework for online parameter personalization of ICDs. Heart states can be inferred from ECG signals from ECG patches, which can be used to create a digital twin of the patient. Reinforcement learning then use the digital twin as environment to explore parameter settings with less misdiagnosis. Experiments were performed on three virtual patients with specific and evolving heart conditions, and the result shows that our proposed approach can identify ICD parameter settings that can achieve better performance compared to default parameter settings.

Keywords: Wearable or portable devices for vital signal monitoring, Physiological monitoring & diagnistic devices - Blood pressure, Clinical engineering - Health technology / system management and assessment
Abstract: Hypertension is a major global cause of morbidity and mortality. Home Blood Pressure Monitoring (HBPM) has the potential to help diagnose patients experiencing isolated nocturnal hypertension who may otherwise be missed. This paper investigates potential diagnostic thresholds for diagnosing isolated nocturnal hypertension using dawn and dusk HBPM measurements in the BP-Eth ambulatory blood pressure monitoring (ABPM) database. Depending on whether European or American diagnostic guidelines for hypertension were used, incidence of isolated nocturnal hypertension in the BP-Eth database was 17.1% or 16.8%, respectively. Using averaged dawn and dusk HBPM measurements to diagnose isolated nocturnal hypertension yielded an AUROC of 0.79 (European guidelines) or 0.84 (American guidelines). The SBP and DBP diagnostic thresholds required to detect 80% of cases of isolated nocturnal hypertension were found to be 125.4 mmHg and 75.7 mmHg, respectively (European guidelines) or 117.6 mmHg and 74.3 mmHg, respectively (American guidelines). These thresholds corresponded to a sensitivity of 80% and specificity of 63% (European guidelines) or sensitivity of 83% and specificity of 65% (American guidelines). These results demonstrate the potential for HBPM to function as an intermediate step in screening patients, determining which patients require more intensive ABPM monitoring for detection of isolated nocturnal hypertension.

Keywords: Models and simulations of therapeutic devices and systems, Cardiovascular assessment and diagnostic technologies
Abstract: Aortic valvuloplasty is a minimally invasive procedure for the dilatation of stenotic aortic valves. Rapid ventricular pacing is an established technique for balloon stabilization during this procedure. However, low cardiac output due to the pacing is one of the inherent risks, which is also associated with several potential complications.

This paper proposes a numerical modelling approach to understand the effect of different inflation levels of a valvuloplasty balloon catheter on the positional instability caused by a pulsating blood flow. An unstretched balloon catheter model was crimped into a tri-folded configuration and then inflated to several levels. Ten different inflation levels were then tested, and a Fluid-Structure Interaction model was built to solve interactions between the balloon and the blood flow modelled in an idealised aortic arch. Our computational results show that the maximum displacement of the balloon catheter increases with the inflation level, with a small step at around 50% inflation and a sharp increase after reaching 85% inflation. This work represents a substantial progress towards the use of simulations to solve the interactions between a balloon catheter and pulsating blood flow.

Keywords: Transdermal drug delivery, Infusion pumps, Robotic-aided therapies - Wireless therapeutic systems
Abstract: Micro Transdermal Interface Platforms (MicroTIPs) will combine minimally invasive microneedle arrays with highly miniaturized sensors, actuators, control electronics, wireless communications and artificial intelligence. These patch-like devices will be capable of autonomous physiological monitoring and transdermal drug delivery, resulting in increased patient adherence and devolved healthcare. In this paper, we experimentally demonstrate the feasibility of controlled transdermal drug delivery using a combination of 500 um tall silicon microneedles, a commercial micropump, pressure and flow sensors, and bespoke electronics. Using ex-vivo human skin samples and a customized application/retraction system, leak-free delivery of volumes ranging from 0.7-1.1 mL has been achieved in under one hour.

Keywords: Image reconstruction and enhancement - Compressed sensing / Sampling, Image reconstruction and enhancement - Machine learning / Deep learning approaches, Machine learning / Deep learning approaches
Abstract: Magnetic resonance imaging (MRI) is a powerful imaging modality that revolutionizes medicine and biology. The imaging speed of high-dimensional MRI is often limited, which constrains its practical utility. Recently, low-rank tensor models have been exploited to enable fast MR imaging with sparse sampling. Most existing methods use some pre-defined sampling design, and active sensing has not been explored for low-rank tensor imaging. In this paper, we introduce an active low-rank tensor model for fast MR imaging. We propose an active sampling method based on a Query-by-Committee model, making use of the benefits of low-rank tensor structure.Numerical experiments on a 3-D MRI data set with Cartesian sampling designs demonstrate the effectiveness of the proposed method.

Keywords: Image reconstruction and enhancement - Compressed sensing / Sampling, Iterative image reconstruction, Image reconstruction and enhancement - Machine learning / Deep learning approaches
Abstract: Magnetic Resonance (MR) Fingerprinting is an emerging transient-state imaging paradigm, which enables the quantization of multiple MR tissue parameters in a single experiment. Balanced steady-state free precession (bSSFP)-based MR Fingerprinting has excellent signal-to-noise characteristics and also allows for acquiring both tissue parameter maps and field inhomogeneity maps. However, field inhomogeneity often results in complex magnetization evolutions in bSSFP-based MR Fingerprinting, which creates significant challenges in image reconstruction. In this paper, we introduce a new method to address the image reconstruction problem. The proposed method incorporates a low-dimensional nonlinear manifold learned from an ensemble of magnetization evolutions using a deep autoencoder. It provides much better representation power than a low-dimensional linear subspace in capturing complex magnetization evolutions. We formulate the image reconstruction problem with this nonlinear model and solve the resulting optimization problem using an algorithm based on variable splitting and the alternating direction method of multipliers. We evaluate the performance of the proposed method using numerical experiments and demonstrate that it significantly outperforms the state-of-art method using a linear subspace model.

Keywords: Image reconstruction and enhancement - Machine learning / Deep learning approaches, Image reconstruction and enhancement - Image synthesis, Image retrieval
Abstract: Capturing high-resolution magnetic resonance(MR) images is a time consuming process, which makes it unsuitable for medical emergencies and pediatric patients. Low-resolution MR imaging, by contrast, is faster than its high-resolution counterpart, but it compromises on fine details necessary for a more precise diagnosis. Super-resolution (SR), when applied to low-resolution MR images, can help increase their utility by synthetically generating high-resolution images with little additional time. In this paper, we present a SR technique for MR images that is based on generative adversarial networks(GANs), which have proven to be quite useful in generating sharp-looking details in SR. We introduce a conditional GAN with perceptual loss, which is conditioned upon the input low-resolution image, which improves the performance for isotropic and anisotropic MRI super-resolution.

Keywords: Image reconstruction and enhancement - Machine learning / Deep learning approaches, Image reconstruction and enhancement - Image synthesis, Image feature extraction
Abstract: Gait is one of the most frequently used forms of human movement during daily activities. The majority of works focus on exploring the dynamic factors during gait. Different from previous works, we adapt an image prediction task for anticipating the next frame in process of gait. In this work, we present a novel framework for human gait plantar pressure prediction using Spatio-temporal Transformer. We train the model to predict the next plantar pressure image in an image series while also learning frame feature encoders that predict the features of subsequent frames in the sequence. We proposed two new components in our loss function for considering temporality as well as smaller values in the image. Our method has the advantage over existing models in that it preserves the sequential sequence of observed images while also preserving long-range dependency, which are both important for the prediction. Our model achieves superior results over several competitive baselines on the CAD WALK database. Clinical relevance— This work can be used in robotic exoskeleton devices which are intelligent systems designed to improve gait performance and quality of life for the wearer that are being used to assist the recovery of walking ability for patients with disorders.

Keywords: Image reconstruction and enhancement - Tomographic reconstruction, Ultrasound imaging - Breast
Abstract: Ultrasound computed tomography (USCT) is considered to have great potential for breast cancer screening. Compared with the ray based methods, the reconstructed image using full waveform inversion (FWI) methods have higher spatial resolution. However, the results of FWI is difficult to converge to the real value when cycle skipping occurs. In this paper, a cross-correlation full waveform inversion (CC-FWI) is proposed for USCT image reconstruction. In the first stage, the ajoint source is adjusted as the residual of predicted signal and time-shifted measured signal to avoid cycle skipping. In the remaining stage, the FWI with source encoding is employed to accelerate convergence. The simulations are conducted to demonstrate the validity of the proposed algorithm. The root mean squared error (RMSE) of the proposed algorithm is much smaller than that of conventional FWI. The results suggest that CC-FWI is effective in avoiding cycle skipping.

Clinical relevance— New imaging modalities of high resolution, safety to examines for early-stage breast cancer imaging are urgently needed for researching and development. Ultrasound computed tomography (USCT) is supposed to meet the above requirements and it can be potentially deployed in breast scanning.

Keywords: Image analysis and classification - Digital Pathology, Image classification, Image feature extraction
Abstract: Preterm infants in a neonatal intensive care unit (NICU) are continuously monitored for their vital signs, such as heart rate and oxygen saturation. Body motion patterns are documented intermittently by clinical observations. Changing motion patterns in preterm infants are associated with maturation and clinical events such as late-onset sepsis and seizures. However, continuous motion monitoring in the NICU setting is not yet performed. Video-based motion monitoring is a promising method due to its non-contact nature and therefore unobtrusiveness. This study aims to determine the feasibility of simple video-based methods for infant body motion detection. We investigated and compared four methods to detect the motion in videos of infants, using two datasets acquired with different types of cameras. The thermal dataset contains 32 hours of annotated videos from 13 infants in open beds. The RGB dataset contains 9 hours of annotated videos from 5 infants in incubators. The compared methods include background substruction (BS), sparse optical flow (SOF), dense optical flow (DOF), and oriented FAST and rotated BRIEF (ORB). The detection performance and computation time were evaluated by the area under receiver operating curves (AUC) and run time. We conducted experiments to detect motion and gross motion respectively. In the thermal dataset, the best performance of both experiments is achieved by BS with mean (standard deviation) AUCs of 0.86 (0.03) and 0.93 (0.03). In the RGB dataset, SOF outperforms the other methods in both experiments with AUCs of 0.82 (0.10) and 0.91 (0.05). All methods are efficient to be integrated into a camera system when using low-resolution thermal cameras.

Keywords: Image analysis and classification - Digital Pathology, Image feature extraction, Optical imaging
Abstract: Meningioma is the most common intracranial tumor in adulthood. With a clear female predominance and a recurrence rate that reaches 20%, it is, despite being considered a benign tumor, a pathology that greatly compromises post-diagnosis quality of life. Its prone to recur or progress to a higher degree is difficult to predict in the absence of obvious histological criteria. This project aims to develop an automatic methodology to aid in the diagnosis of meningiomas that is objective and easily reproducible. The methodology is based on histopathological image analysis using artificial intelligence and machine learning algorithms. It includes a semi-automatic process of identification and cleaning of the scanned samples, an automatic detection of the nuclei of each image and, finally, the parameterization of the samples. The obtained data together with the clinical information will be analyzed using statistical methods in order to provide a methodology to support clinical diagnosis and decision-making in patient management. The result is the development of an effective methodology that generates a set of data associated with morphological parameters with different trends according to the pathological groups studied. A tool has been developed that allows an effective semiautomatic analysis of the images to evaluate these parameters in an objective and reproducible way, helping in clinical decision-making and facilitating to undertake projects with large sample series.

Keywords: Image analysis and classification - Digital Pathology, Image analysis and classification - Machine learning / Deep learning approaches, Image feature extraction
Abstract: Whole Slide Images (WSIs) in digital pathology are used to diagnose cancer subtypes. The difference in procedures to acquire WSIs at various trial sites gives rise to variability in the histopathology images, thus making consistent diagnosis challenging. These differences may stem from variability in image acquisition through multi-vendor scanners, variable acquisition parameters, and differences in staining procedure; as well, patient demographics may bias the glass slide batches before image acquisition. These variabilities are assumed to cause a domain shift in the images of different hospitals. It is crucial to overcome this domain shift because an ideal machine-learning model must be able to work on the diverse sources of images, independent of the acquisition center. A domain generalization technique is leveraged in this study to improve the generalization capability of a Deep Neural Network (DNN), to an unseen histopathology image set (i.e., from an unseen hospital/trial site) in the presence of domain shift. According to experimental results, the conventional supervised-learning regime generalizes poorly to data collected from different hospitals. However, the proposed hospital-agnostic learning can improve the generalization considering the low-dimensional latent space representation visualization, and classification accuracy results.

Keywords: Image analysis and classification - Digital Pathology, Image analysis and classification - Machine learning / Deep learning approaches, Image classification
Abstract: Transfer learning from ImageNet pretrained weights is widely used when training Deep Learning models on a Histopathology dataset. However, the visual features of the two domains are different. Rather than ImageNet pretrained weights, pre-training on a Histopathology dataset may provide better initialization. To prove this hypothesis, we train two commonly used Deep Learning model architectures - ResNet and DenseNet on a complex Histopathology classification dataset, and compare transfer learning performance with ImageNet pretrained weights. Based on the fine-tuning on three histopathology datasets including two different stains (H&E and IHC), we show that the domain specific pretrained weights are better suited for transfer learning. This is reflected by higher performance, lower training time as well as better feature reuse.

Keywords: Image analysis and classification - Digital Pathology, Optical imaging and microscopy - Fluorescence microscopy, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Multiplexed immunofluorescence provides an unprecedented opportunity for studying specific cell-to-cell and cell microenvironment interactions. We employ graph neural networks to combine features obtained from tissue morphology with measurements of protein expression to profile the tumour microenvironment associated with different tumour stages. Our framework presents a new approach to analysing and processing these complex multi-dimensional datasets that overcomes some of the key challenges in analysing these data and opens up the opportunity to abstract biologically meaningful interactions.

Keywords: Image analysis and classification - Digital Pathology, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Microsatellite instability (MSI) is a clinically important characteristic of colorectal cancer. Standard diagnosis of MSI is performed via genetic analyses, however these tests are not always included in routine care. Histopathology whole-slide images (WSIs) are the gold-standard for colorectal cancer diagnosis and are routinely collected. This study develops a model to predict MSI directly from WSIs. Making use of both weakly- and self-supervised deep learning techniques, the proposed model shows improved performance over conventional deep learning models. Additionally, the proposed framework allows for visual interpretation of model decisions. These results are validated in internal and external testing datasets.

Keywords: Neural interfaces - Tissue-electrode interface, Neural interfaces - Biomaterials, Neural interfaces - Bioelectric sensors
Abstract: The biological response to electrodes implanted in the brain has been a long-standing barrier to achieving a stable tissue device-interface. Understanding the mechanisms underlying this response could explain phenomena including recording instability and loss, shifting stimulation thresholds, off-target effects of neuromodulation, and stimulation-induced depression of neural excitability. Our prior work detected differential expression in hundreds of genes following device implantation. Here, we extend upon that work by providing new analyses using differential co-expression analysis, which identifies changes in the correlation structure between groups of genes detected at the interface in comparison to control tissues. We used an “eigengene” approach to identify hub genes associated with each module. Our work adds to a growing body of literature which applies new techniques in molecular biology and computational analysis to long-standing issues surrounding electrode integration with the brain.

Keywords: Neural stimulation, Neuromuscular systems - EMG models, Neuromuscular systems - Central mechanisms
Abstract: Recent studies have reported that transcutaneous spinal stimulation (tSCS) may facilitate improved upper limb motor function in those with incomplete tetraplesia. However, little is known about how tSCS engages upper limb motor pools. This study aimed to explore the extent to which discrete upper limb motor pools can be selectively engaged via altering stimulus location and intensity. 14 participants with intact nervous systems completed two test visits, during which posterior root-muscle reflexes (PRMR) were evoked via a 3x3 cathode matrix applied over the cervicothoracic spine. An incremental recruitment curve at C7 vertebral level was initially performed to attain minimal threshold intensity (MTI) in each muscle. Paired pulses (1ms square monophasic with inter-pulse interval of 50ms) were subsequently delivered at a frequency of 0.25Hz at two intensities (MTI and MTI+20%) across all nine locations. in a random order. Evoked response to the 1st (PRMR1) and 2nd (PRMR2) stimuli were recorded from four upper limb muscles. A significant effect of spinal level was observed in all muscles for PRMR1 with greater responses recorded more caudally. Unexpectedly, contralateral cathode placement significantly increased PRMR1 in Biceps Brachii (P=0.012), Flexor Carpi Radialis (P=0.035) and Abductor Pollicis Brevis (P=0.001). Post-activation depression (PAD) was also significantly increased with contralateral cathode placement in Biceps Brachii (P=0.001), Triceps Brachii (P=0.012) and Flexor Carpi Radialis (P=0.0001). These results suggest that some level of unilateral motor pool selectivity may be attained via altering stimulus intensity and location during cervicothoracic tSCS.

Keywords: Neural interfaces - Implantable systems, Neural stimulation, Neural interfaces - Microelectrode technology
Abstract: Abstract— Optogenetics is a powerful neuroscientific tool which allows neurons to be modulated by optical stimulation. Despite widespread optogenetic experimentation in small animal models, optogenetics in non-human primates (NHPs) remains a niche field, particularly at the large scales necessary for multi-regional neural research. We previously published a large-scale, chronic optogenetic cortical interface for NHPs which was successful but came with a number of limitations. In this work, we present an optimized interface which improves upon the stability and scale of our previous interface while using more easily replicable methods to increase our system’s availability to the scientific community. Specifically, we (1) demonstrate the long-term (~3 months) optical access to the brain achievable using a commercially-available transparent artificial dura with embedded electrodes, (2) showcase large-scale optogenetic expression achievable with simplified (magnetic resonance-free) surgical techniques, and (3) effectively modulated the expressing areas at large scales (~1 cm2) by light emitting diode (LED) arrays assembled in-house.

Keywords: Neurological disorders - Stroke, Neural stimulation, Neural signal processing
Abstract: Brain stimulation has emerged as a novel therapy for ischemic stroke, a major cause of brain injury that often results in lifelong disability. Although past works in rodents have demonstrated protective effects of stimulation following stroke, few of these results have been replicated in humans due to the anatomical differences between rodent and human brains and a limited understanding of stimulation-induced network changes. Therefore, we combined electrophysiology and histology to study the neuroprotective mechanisms of electrical stimulation following cortical ischemic stroke in non-human primates. To produce controlled focal lesions, we used the photothrombotic method to induce targeted vasculature damage in the sensorimotor cortices of two macaques while collecting electrocorticography (ECoG) signals bilaterally. In another two monkeys, we followed the same lesioning procedures and applied repeated electrical stimulation via an ECoG electrode adjacent to the lesion. We studied the protective effects of stimulation on neural dynamics using ECoG signal power and coherence. In addition, we performed histological analysis to evaluate the differences in lesion volume. In comparison to controls, the ECoG signals showed decreased gamma power across the sensorimotor cortex in stimulated animals. Meanwhile, Nissl staining revealed smaller lesion volumes for the stimulated group, suggesting that electrical stimulation may exert neuroprotection by suppressing post-ischemic neural activity. With the similarity between NHP and human brains, this study paves the path for developing effective stimulation-based therapy for acute stroke in clinical studies.

Keywords: Neural stimulation, Neural interfaces - Implantable systems
Abstract: Polycarbonate is a polymer that has been widely used including medical application due to its useful properties. It has high temperature resistance, biocompatibility, transparency and low water absorption rate, which are needful characteristics for packaging material of implantable neural prosthetic devices. In this study, we investigated fabrication of neural electrode with polycarbonate film using standard photolithography process and heated hydraulic press for thermal lamination. First, oxygen plasma surface treatment was performed to increase the adhesion between metal and polycarbonate film. Then thin layer of titanium and gold layer were deposited. Metal layer is patterned through standard photolithography techniques. After completing the metal patterning, thermal lamination was performed with site opened polycarbonate film.

Keywords: Neural interfaces - Implantable systems, Neural stimulation, Brain physiology and modeling - Neural dynamics and computation
Abstract: Biphasic pulsatile stimulation is the present standard for neural prosthetic use, and it is used to understand connectivity and functionality of the brain in brain mapping studies. While pulses have been shown to drive behavioral changes, such as biasing decision making, they have deficits. For example, cochlear implants restore hearing but lack the ability to restore pitch perception. Recent work shows that pulses produce artificial synchrony in networks of neurons and non-linear changes in firing rate with pulse amplitude. Studies also show galvanic stimulation, delivery of current for extended periods of time, produces more naturalistic behavioral responses than pulses. In this paper, we use a winner-take-all decision-making network model to investigate differences between pulsatile and galvanic stimulation at the single neuron and network level while accurately modeling the effects of pulses on neurons for the first time. Results show pulses bias spike timing and make neurons more resistive to natural network inputs than galvanic stimulation at an equivalent current amplitude.

Keywords: Brain-computer/machine interface, Neural signals - Machine learning & Classification, Human performance - Speech
Abstract: Speech decoding from brain activity can enable development of brain-computer interfaces (BCIs) to restore naturalistic communication in paralyzed patients. Previous work has focused on development of decoding models from isolated speech data with a clean background and multiple repetitions of the material. In this study, we describe a novel approach to speech decoding that relies on a generative adversarial neural network (GAN) to reconstruct speech from brain data recorded during a naturalistic speech listening task (watching a movie). We compared the GAN-based approach, where reconstruction was done from the compressed latent representation of sound decoded from the brain, with several baseline models that reconstructed sound spectrogram directly. We show that the novel approach provides more accurate reconstructions compared to the baselines. These results underscore the potential of GAN models for speech decoding in naturalistic noisy environments and further advancing of BCIs for naturalistic communication.

Keywords: Neural signal processing, Brain-computer/machine interface, Neural interfaces - Implantable systems
Abstract: Virtual reality (VR) offers a robust platform for human behavioral neuroscience, granting unprecedented experimental control over every aspect of an immersive and interactive visual environment. VR experiments have already integrated non-invasive neural recording modalities such as EEG and functional MRI to explore the neural correlates of human behavior and cognition. Integration with implanted electrodes would enable significant increase in spatial and temporal resolution of recorded neural signals and the option of direct brain stimulation for neurofeedback. In this paper, we discuss the first such implementation of a VR platform with implanted electrocorticography (ECoG) and stereo-electroencephalography (sEEG) electrodes in human, in-patient subjects. Noise analyses were performed to evaluate the effect of the VR headset on neural data collected in two VR-naïve subjects, one child and one adult, including both ECOG and sEEG electrodes. Results demonstrate an increase in line noise power (57-63Hz) while wearing the VR headset that is mitigated effectively by common average referencing (CAR), and no significant change in the noise floor bandpower (125-240Hz). To our knowledge, this study represents first demonstrations of VR immersion during invasive neural recording with in-patient human subjects.

Keywords: Neural interfaces - Microelectrode technology, Neural interfaces - Tissue-electrode interface, Brain physiology and modeling - Neural circuits
Abstract: A striking example of the brain’s complexity and continued plasticity is the addition of new neuronal components to a circuit in a process called neurogenesis. Two brain regions exhibit profound circuit remodeling through this process – the olfactory bulb and hippocampus. However, how local network changes in both regions influence global circuit rewiring and dynamic network features remain largely unexplored due to the lack of spatiotemporal resolution technology and large-scale electrophysiological activity recordings. Here, we demonstrate large-scale recordings using a high-density neurochip to reveal multimodal circuit-wide electrophysiological properties and layer-specific functional connectivity in the olfactory bulb and hippocampal networks. Our findings illustrate simultaneous recordings from the entire network, which allows us to quantify synchronous electrophysiological parameter differences and layer-specific waveform markers. Examining pairwise cross-covariance between active electrode pairs reveals individual neuronal ensemble contributions to synchronous activation between layers and hub microcircuits, demonstrating network-wide rewiring. Our study suggests a novel tool to address the computational implications of large-scale activity patterns in functional multimodal neurogenic circuits.

Keywords: Neural stimulation - Deep brain, Motor learning, neural control, and neuromuscular systems, Brain physiology and modeling - Sensory-motor
Abstract: Traditional methods to access subcortical structures involve the use of anatomical atlases and high precision stereotaxic frames but suffer from significant variations in implantation accuracy. Here, we leveraged the use of the ROSA One(R) Robot Assistance Platform in non-human primates to study electrophysiological interactions of the corticospinal tract with spinal cord circuits. We were able to target and stimulate the corticospinal tract within the internal capsule with high accuracy and efficiency while recording spinal local field potentials and multi-unit spikes. Our method can be extended to any subcortical structure and allows implantation of multiple deep brain stimulation probes at the same time.

Keywords: Brain functional imaging - Source localization, Brain functional imaging - EEG
Abstract: Electrophysiological brain source localization consists in estimating the source positions and activities responsible for the (S)EEG measurements. The localization procedure is usually carried out in the time domain, however in specific situations the activities of interest can be located at well defined frequencies, e.g. in response to a rhythmic stimulation. This paper addresses the problem of sparse localization of multiple sources oscillating at the same frequency. In particular the non-unicity of the solution is emphasized, as alternative source maps involving equivalent or less number of sources can be found, challenging source localization methods based on sparsity. These limitations are illustrated under a realistic SEEG simulation framework, and the usefulness to perform localization for this modality is strengthen out.

Keywords: Brain-computer/machine interface, Brain physiology and modeling - Sensory-motor, Neuromuscular systems - EMG processing and applications
Abstract: A better understanding of reward signaling in the sensorimotor cortices can aid in developing Reinforcement Learning-based Brain-Computer Interfaces (RLBCI) for restoration of movement functions with fewer implants. Brain-computer interfaces (BCIs) using local field potentials (LFPs) have recently achieved performance comparable to spike-BCIs [1]. With superior stability over time, LFPs may be the preferred signal for BCIs. We show that sensorimotor LFPs can provide reward level information (R1 – R3) like spikes[2]. We used a cued reward-level reaching task in which reward information was temporally dissociated from movement information. This allowed the study of reward- and movement-related modulations in LFPs. We recorded simultaneously from contralateral primary -somatosensory (S1), -motor (M1), and the dorsal premotor (PMd) cortices in a female Macaca Mulatta. We found that all three cortices’ average beta band (14-30 Hz) amplitude showed robust modulation with reward levels during the cue presentation period. Such modulation was consistently observed after controlling for cue color, differences in behavioral variables, and electromyogram (EMG) activity. Statistical amplitude analysis showed that reward level could be extracted from the simple LFP feature of beta band amplitude, even before a reaching target appeared, and no specific reach plan could be developed.

Keywords: Bio-electric sensors - Sensing methods, Physiological monitoring - Instrumentation, Bio-electric sensors - Sensor systems
Abstract: Dry-contact electrodes are increasingly being used for EEG recordings in both research studies and consumer products. They are more user-friendly and better suited for long-term recordings. However, dry-contact electrodes also bring challenges with respect to the stability and impedance of the electrode-skin interface. We propose a methodology to characterize and compare dry-contact electrodes. The characterization is based on measuring the electrode-skin impedance spectrum, fit a parametric model of the electrode-skin interface to the measured spectrum, and calculate the resulting thermal noise spectrum. Thereby it is possible to relate the noise of an EEG recording to the theoretical noise contribution from the electrode-skin interface. To demonstrate the methodology, we performed an empirical study comparing two types of dry-contact electrodes in an ear-EEG setup. The electrodes were IrO₂, previously used for ear-EEG, and a new design based on Ag/AgCl. Here, we related the noise floor of an auditory steady-state response (ASSR) to the thermal noise spectrum of the electrode-skin interface. The study showed similar impedance and EEG recording quality for the two electrode types, and the thermal noise of the electrode-skin interface was below the noise floor of the EEG recordings for both electrode types.

Keywords: Bio-electric sensors - Sensor systems, Physiological monitoring - Instrumentation, Health monitoring applications
Abstract: Fetal electrocardiography (fECG) has gotten widespread interest in the last years as technology for fetal monitoring. Compared to cardiotocography (CTG), the current state of the art, it can be designed in smaller formfactor and is thus suited for long-term and unsupervised monitoring.

In the present study we evaluated a wearable system which is based on CSEM’s cooperative sensors, a versatile technology that allows for the measurement of multiple biosignals and an easy integration into a garment or patch. The system was tested on 25 patients with singleton pregnancies and an age of gestation ≥ 37 weeks. To reject unreliable fetal heart rate (fHR) estimations, the signal processing algorithm provides a signal quality index.

In 12 out of 21 patients available for analysis, a good performance of fHR estimations was obtained with a mean absolute error < 5 bpm and an acceptance rate >70%. However, the remaining 9 patients showed low acceptance rates and high errors. Besides investigating the source of these high errors, future work includes the investigating improved signal processing algorithms, different body positions and the use of dry electrodes.

Keywords: Wearable sensor systems - User centered design and applications, Physiological monitoring - Novel methods
Abstract: High quality sleep monitoring is done using EEG electrodes placed on the skin. This has traditionally required assistance by an expert when the equipment needed to mounted. However, this creates a limitation in how cheap and easy it can be to record sleep in the subject's own home.

Here we present a data set of 120 home recordings of sleep, in which subjects use self-applied ear-EEG monitoring equipment.

We compare this data set to a previously recorded data set with both ear-EEG and polysomnography, which was applied by an expert.

Clinical relevance: On all tested metrics, self applied sleep recordings behaved the same as expert applied. This indicates that ear-EEG can reliably be used as a home sleep monitor, even when subjects apply the equipment themselves.

Keywords: Wearable low power, wireless sensing methods, Bio-electric sensors - Sensor systems, IoT sensors for health monitoring
Abstract: In recent years, in-ear electroencephalography(EEG) was demonstrated to record signals of similar quality compared to standard scalp-based EEG, and clinical applications of objective hearing threshold estimations have been reported. Existing devices, however, still lack important features. In fact, most of the available solutions are based on wet electrodes, require to be connected to external acquisition platforms, or do not offer on-board processing capabilities. Here we overcome all these limitations, presenting an ear- EEG system based on dry electrodes that includes all the acquisition, processing, and connectivity electronics directly in the ear bud. The earpiece is equipped with an ultra-low power analog front-end for analog-to-digital conversion, a low-power MEMS microphone, a low-power inertial measurement unit,and an ARM Cortex-M4 based microcontroller enabling on- board processing and Bluetooth Low Energy connectivity. The system can stream raw EEG data or perform data processing directly in-ear. We test the device by analysing its capability to detect brain response to external auditory stimuli, achieving 4 and 1.3 mW power consumption for data streaming or on board processing, respectively. The latter allows for 600 hours operation on a PR44 zinc-air battery. To the best of our knowledge, this is the first wireless and fully self-contained ear-EEG system performing on-board processing, all embedded in a single earbud.

Keywords: Physiological monitoring - Instrumentation, Optical and photonic sensors and systems, Physiological monitoring - Modeling and analysis
Abstract: PhotoPlethysmoGraphy (PPG) is ubiquitously employed in wearable devices for health monitoring. Photodiode signal inversion is observed in rare occasions, most of the time when the sensor is pressed against the skin. We report in this article such observations made at the right common carotid artery site. Indeed we have systematically observed a photodiode signal inversion when the PPG sensor is placed where the pulse is the best felt at the carotid. In addition to be inverted, the pulse is steeper during the systolic phase. Such inversion has implications in terms of pulse arrival time (PAT) measurements In our experiments, this causes a difference of 20 ms in the carotid PAT when measured at the absolute maximum slope. The mechanical and optical properties of tissues must be better accounted to explain the PPG signal morphology.

Clinical Relevance— Understanding the role of mechanical tissue properties seems relevant in order to obtain more reproducible results in PPG signal analysis.

Keywords: Wearable wireless sensors, motes and systems, Physiological monitoring - Modeling and analysis, Health monitoring applications
Abstract: Stress is has been classified as the health epidemic of the 21st century with an increasingly active research interest within the fields of psychology, neuroscience, medicine, and more recently affective computing. At present, stress is identified through cortisol levels in saliva but there is no unanimously accepted standard for continuous stress evaluation. With recent development in wearable sensors, many scientist are interested in stress identification through physiological signals such as the Heart rate variability (HRV). In this paper, we present a novel supervised machine learning-based algorithm to detect stress from HRV derived from both electrocardiograms (ECG) and photoplethysmograms (PPG). HRV features from ECG and PPG signals of 46 healthy subjects were analysed and used to separately train and test a Random Forest algorithm. In both datasets, stress was accurately identified with more than 80% F1-score and 90% AUC. Results show that PPG is a good surrogate to ECG for HRV analysis and stress detection. The proposed algorithm has the potential to assist researchers and clinicians in the automated continuous analysis of stress.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Data mining and big data methods - Inter-subject variability and personalized approaches, Physiological systems modeling - Multivariate signal processing
Abstract: Electroencephalography (EEG) is shown to be a valuable data source for evaluating subjects' mental states. However, the interpretation of multi-modal EEG signals is challenging, as they suffer from poor signal-to-noise-ratio, are highly subject-dependent, and are bound to the equipment and experimental setup used, (i.e. domain). This leads to machine learning models often suffer from poor generalization ability, where they perform significantly worse on real-world data than on the exploited training data. Recent research heavily focuses on cross-subject and cross-session transfer learning frameworks to reduce domain calibration efforts for EEG signals. We argue that multi-source learning via learning domain-invariant representations from multiple data-sources is a viable alternative, as the available data from different EEG data-source domains (e.g., subjects, sessions, experimental setups) grow massively. We propose an adversarial inference approach to learn data-source invariant representations in this context, enabling multi-source learning for EEG-based brain-computer interfaces. We unify EEG recordings from different source domains (i.e., emotion recognition datasets SEED, SEED-IV, DEAP, DREAMER), and demonstrate the feasibility of our invariant representation learning approach in suppressing data-source-relevant information leakage by 35% while still achieving stable EEG-based emotion classification performance.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: We provide a regularization framework for subject transfer learning in which we train an encoder and classifier to minimize classification loss, subject to a penalty measuring independence between the latent representation and the subject label. We introduce three notions of independence and corresponding penalty terms using mutual information or divergence as a proxy for independence. For each penalty term, we provide several concrete estimation algorithms, using analytic methods as well as neural critic functions. We provide a hands-off strategy for applying this diverse family of regularization algorithms to a new dataset, which we call ``AutoTransfer". We evaluate the performance of these individual regularization strategies and our AutoTransfer method on EEG, EMG, and ECoG datasets, showing that these approaches can improve subject transfer learning for challenging real-world datasets.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Pattern recognition, Signal pattern classification
Abstract: Attention-deficit hyperactivity disorder (ADHD) affects at least 5% of the world population and can disturb normal development causing serious issues in adulthood. Therefore, it is important to develop tools to help detecting ADHD so that treatment can start as soon as possible. Plus, the differentiation of ADHD in its subtypes is important to define the recommended treatment. Here we present original research to investigate the hypothesis of using a Spiking Neural Networks (SNN) EEG signals classifier for automated diagnostic of ADHD subtypes. This research used data from 243 patients and healthy volunteers acquired as part of the Healthy Brain Network. These resting state EEG signals were collected from 5-minutes scan with a 128 channel 500 Hz system. For benchmarking, we present a comparison of the SNN performance with a support vector machine, a k-nearest neighborhood, a random forest algorithm and a multi-layer perceptron. We present experiments for both the diagnostics of ADHD and for detecting which ADHD subtype the patient has. SNN presented a 72.00% accuracy for detecting ADHD surpassing all the other techniques by 9.1% and 68% in detecting if the subject is a member of the Combined ADHD, Inattentive ADHD or control groups (18% better than the second-best technique).

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Data mining and big data methods - Biosignal classification, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: Among the different modalities to assess emotion, electroencephalogram (EEG), representing the electrical brain activity, achieved motivating results over the last decade. Emotion estimation from EEG could help in the diagnosis or rehabilitation of certain diseases. In this paper, we propose a dual model considering two different representations of EEG feature maps: 1) a sequential based representation of EEG band power, 2) an image-based representation of the feature vectors. We also propose an innovative method to combine the information based on a saliency analysis of the image-based model to promote joint learning of both model parts. The model has been evaluated on four publicly available datasets: SEED-IV, SEED, DEAP and MPED. The achieved results outperform results from state-of-the-art approaches for three of the proposed datasets with a lower standard deviation that reflects higher stability. For sake of reproducibility, the codes and models proposed in this paper are available at https://github.com/VDelv/Emotion-EEG.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: Alzheimer’s Disease (AD) is the most common form of dementia. Mild Cognitive Impairment (MCI) is the term given to the stage describing prodromal AD and represents a ‘risk factor’ in early-stage AD diagnosis from normal cognitive decline due to ageing. The electroencephalogram (EEG) has been studied extensively for AD characterization, but reliable early-stage diagnosis continues to present a challenge. The aim of this study was to introduce a novel way of classifying between AD patients, MCI subjects, and age-matched healthy control (HC) subjects using EEG-derived feature images and deep learning techniques. The EEG recordings of 141 age-matched subjects (52 AD, 37 MCI, 52 HC) were converted into 2D greyscale images representing the Pearson correlation coefficients and the distance Lempel-Ziv Complexity (dLZC) between the 21 EEG channels. Each feature type was computed from EEG epochs of 1s, 2s, 5s and 10s segmented from the original recording. The CNN architecture AlexNet was modified and employed for this three-way classification task and a 70/30 split was used for training and validation with each of the different epoch lengths and EEG-derived images. Whilst a maximum classification accuracy of 73.49% was obtained using dLZC-derived images from 10s epochs as input to the model, the classification accuracy reached 98.13% using the images obtained from Pearson correlation coefficients and 5s epochs.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: Alzheimer's disease (AD) is the main cause of dementia and Mild cognitive impairment (MCI) is a prodromal stage of AD whose early detection is considered crucial as it can contribute in slowing the progression of AD. In our study we attempted to classify a subject into AD, MCI, or Healthy Control (HC) groups with the use of electroencephalogram (EEG) data. Due to the time-series nature of EEG we experimented with the powerful recurrent neural network (RNN) classifiers and more specifically with models including basic or bidirectional Long Short-Term Memory (LSTM) modules. The EEG signals from 17 channels were preprocessed using a 0.1-32 Hz band-pass filter and then segmented into 2-second epochs during which, the subject had closed eyes. Finally, on each segment Fast Fourier Transform (FFT) was applied. To evaluate our models we studied four different classification problems: problem 1: separating subject into three classes (HC, MCI, AD) and problems 2-4: pairwise classifications AD vs. MCI, AD vs. HC and MCI vs. HC. For each problem we employed two different cross-validation approaches (a) by segment and (b) by patient. In the first one, segments from a subject EEG may exist in both training and validations set, while in the second one, all the EEG segments of a subject can only exist in either the training or the validation set. In the AD-MCI-HC classification we achieved an accuracy of 99% by segment cross-validation, which was an improvement to earlier studies that utilized recurrent neural network models. In the pairwise classification problems we achieved over 90% accuracy by segment and over 80% by subject.

Keywords: Data mining and big data methods - Inter-subject variability and personalized approaches, Data mining and big data methods - Pattern recognition, Signal pattern classification
Abstract: The purpose of this study it to assess the effect of sequential learning of self-training support vector machine (ST-S3VM) on short- and long-term surface electromyogram (sEMG) datasets. A machine learning-based supervised classifier is enabling stable, complex, and high-performance motion control. Unlabeled sEMG measurements are easy by the development of wearable sensing technology. Thus, semi-supervised learning methods are attracted attention to utilize unlabeled sEMG data for supervised classifier with a small amount of labeled data. To evaluate robustness of ST-S3VM in realistic conditions, two public datasets which respectively contain a short- and long-term dataset were used. We compared the performance of ST-S3VM with four-kinds of SVM classifiers. In both short- and long-term situations, ST combined classifiers (ST-SVM and ST-S3VM) showed higher performances than the methods without ST (SVM and S3VM). In some cases, ST-S3VM had the best performance, but in other cases, ST-SVM had better performance than ST-S3VM. In order to make better use of unlabeled data, we will develop ST-S3VM to reduce the impact of harmful unlabeled data.

Keywords: Signal pattern classification
Abstract: Within state-of-the-art gesture-based upper-limb myoelectric prosthesis control, gesture recognition commonly relies on the classification of features extracted from electromyographic (EMG) data gathered from the amputee’s residual forearm musculature. Despite best efforts in broadly maximizing gesture recognition accuracy, there does not yet exist a feature-classifier combination accepted as best-practice. In turn, this work hypothesizes that no single feature-classifier combination can consistently maximize accuracy across subjects, positing instead that control schemes should be personalized to the individual. To investigate this hypothesis, the study employed the 40-subject, 49-gesture Ninapro Database 2 (DB2) to compare the performance of 7 different historic, more recent, and state-of-the-art feature sets when classified by 5 machine learning algorithms commonly seen within EMG-based pattern recognition literature. The results demonstrate the ability of Linear Discriminant Analysis (LDA) to marginally exceed other more computationally intensive classifiers in terms of mean accuracy, while the feature set which maximized the highest proportion of individuals’ accuracies was shown to vary with both classifier choice and gesture count.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification, Physiological systems modeling - Closed loop systems
Abstract: Accurate identification of the intended hand movement from the surface Electromyography (sEMG) data is desired for effective control of myoelectric lower arm prostheses. This study improves the classification accuracy of hand gestures by using feature arrays, Kalman filter (KF), and a Softmax classifier. We use data from BioPatRec database to classify ten hand movements performed by 17 participants. The proposed classifier achieved 95.3% accuracy without KF, and 99.3% accuracy when KF was used to smooth the training data.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification
Abstract: The collection of Parkinson's Disease (PD) time-series data usually results in imbalanced and incomplete datasets due to the geometric distribution of PD complications' severity scores. Consequently, when training deep convolutional models on these datasets, the models suffer from overfitting and lack generalizability to unseen data. In this paper, we investigated a new framework of Conditional Generative Adversarial Networks (cGANs) as a solution to improve the extrapolation and generalizability of the regression models in such datasets. We used a real-world PD dataset to estimate Dyskinesia severity in patients with PD. The developed cGAN demonstrated significantly better generalizability to unseen data samples than a traditional Convolutional Neural Network with an improvement of 34%. This solution can be applied in similar imbalanced time-series data, especially in the healthcare domain, where balanced and uniformly distributed data samples are not readily available.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Signal pattern classification, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Physical activity recognition in patients with Parkinson’s Disease (PwPD) is challenging due to the lack of large-enough and good quality motion data for PwPD. A common approach to this obstacle involves the use of models trained on better quality data from healthy patients. Models can struggle to generalize across these domains due to motor complications affecting the movement patterns in PwPD and differences in sensor axes orientations between data. In this paper, we investigated the generalizability of a deep convolutional neural network (CNN) model trained on a young, healthy population to PD, and the role of data augmentation on alleviating sensor position variability. We used two publicly available healthy datasets - PAMAP2 and MHEALTH. Both datasets had sensor placements on the chest, wrist, and ankle with 9 and 10 subjects, respectively. A private PD dataset was utilized as well. The proposed CNN model was trained on PAMAP2 in k-fold cross-validation based on the number of subjects, with and without data augmentation, and tested directly on MHEALTH and PD data. Without data augmentation, the trained model resulted in 48.16% accuracy on MHEALTH and 0% on the PD data when directly applied with no model adaptation techniques. With data augmentation, the accuracies improved to 87.43% and 44.78%, respectively, indicating that the method compensated for the potential sensor placement variations between data.

Keywords: Principal component analysis
Abstract: Hand prehension requires a highly coordinated control of contact forces. The high dimensional sensorimotor system of the human hand although operates at ease, poses several challenges when replicated for prosthetic control. This study investigates how the dynamical synergies, coordinated spatial patterns of contact forces, contribute to the contact forces in a grasp, and whether the dynamical synergies could potentially serve as candidates for feedforward and feedback mechanisms. Ten right-handed subjects were recruited to grasp and hold mass-varied objects. The contact forces during this multidigit prehension were recorded using an instrumented grip glove. The dynamical synergies were derived using principal component analysis (PCA). The contact force patterns during the grasps were reconstructed using the first few synergies. The significance of the dynamical synergies and the current challenges and possible applications of the dynamical synergies were discussed along with the integration of the dynamical synergies into prosthetics and exoskeletons that can possibly enable near-natural control.

Keywords: General and theoretical informatics - Machine learning, General and theoretical informatics - Predictive analytics, Imaging Informatics - Image analysis, processing and classification
Abstract: Identifying gene mutation is essential to prognosis and therapeutic decisions for acute myeloid leukemia (AML) but the current gene analysis is inefficient and non-scalable. Pathological images are readily accessible and can be effectively modeled using deep learning. This work aims at predicting gene mutation directly by modeling bone marrow smear images. Traditionally, bone marrow smear slides are cropped into patches with manual segmentation for patch-level modeling. Slide-level modeling, such as multi-instance learning, could aggregate patches for holistic modeling, though suffer from excessive redundancy. In this study, we propose a discriminative multi-instance approach to select useful patches in a coarse-to-fine process. Specifically, we preprocess a slide into patches by using a trained pre-selector network. Then, we rule out low quality patches in the coarse selection with known prior knowledge, and refine the model using gene-discriminative patches in the fine selection. We evaluate the framework for CEBPA, FLT3, and NPM1 gene mutation prediction and obtain 71.67%, 56.26%, and 56.34% F1-score. Further analysis show the effect of different selection criteria on prediction gene mutations using pathological images.

Keywords: Imaging Informatics - Computational pathology, Imaging Informatics - Biomedical imaging marker extraction, Imaging Informatics - Image analysis, processing and classification
Abstract: Eosinophilic esophagitis (EoE) is an allergic inflammatory condition of the esophagus associated with elevated numbers of eosinophils. Disease diagnosis and monitoring require determining the concentration of eosinophils in esophageal biopsies, a time-consuming, tedious and somewhat subjective task currently performed by pathologists. Here, we developed a machine learning pipeline to identify, quantitate and diagnose EoE patients' at the whole slide image level. We propose a platform that combines multi-label segmentation deep network decision support system with dynamics convolution that is able to process whole biopsy slide. Our network is able to segment both intact and not-intact eosinophils with a mean intersection over union (mIoU) of 0.93. This segmentation enables the local quantification of intact eosinophils with a mean absolute error of 0.611 eosinophils. We examined a cohort of 1066 whole slide images from 400 patients derived from multiple institutions. Using this set, our model achieved a global accuracy of 94.75%, sensitivity of 94.13%, and specificity of 95.25% in reporting EoE disease activity. Our work provides state-of-the-art performances on the largest EoE cohort to date, and successfully addresses two of the main challenges in EoE diagnostics and digital pathology, the need to detect several types of small features simultaneously, and the ability to analyze whole slides efficiently. Our results pave the way for an automated diagnosis of EoE and can be utilized for other conditions with similar challenges.

Keywords: Imaging Informatics - Image analysis, processing and classification, Health Informatics - Computer-aided decision making, Imaging Informatics - Biomedical imaging marker extraction
Abstract: Intelligent computer-aided algorithms analyzing photographs of various mouth regions can help in reducing the high subjectivity in human assessment of oral lesions. Very often, in the images, a ruler is placed near a suspected lesion to indicate its location and as a physical size reference. In this paper, we compared two deep-learning networks: ResNeSt and ViT, to automatically identify ruler images. Even though the ImageNet 1K dataset contains a “ruler” class label, the pre-trained models showed low sensitivity. After fine-tuning with our data, the two networks achieved high performance on our test set as well as a hold-out test set from a different provider. Heatmaps generated using three saliency methods: GradCam and XRAI for ResNeSt model, and Attention Rollout for ViT model, demonstrate the effectiveness of our technique.

Keywords: Imaging Informatics - Image registration, segmentation, and compression, Bioinformatics - Bioinformatics for health monitoring, Health Informatics - Informatics for chronic disease management
Abstract: The retina is a unique tissue that extends the human brain in transmitting the incoming light into neural spikes. Researchers collaborating with domain experts proposed numerous deep networks to extract vessels from the retina; however, these techniques have the least response for micro-vessels. The proposed method has developed a stacked ensemble network approach with deep neural architectures for precise vessel extraction. Our method has used bi-directional LSTM for filling gaps in dis-joint vessels and applied W-Net for boundary refinement and emphasizing local regions to achieve better results for micro-vessels extraction. The platform has combined the strength of various networks to improve the automated screening process and has shown promising results on benchmark datasets.

Keywords: Imaging Informatics - Radiomics, General and theoretical informatics - Machine learning
Abstract: MRI-based radiomic models have shown promises in predicting the response to neoadjuvant chemotherapy in breast cancer. However, it is difficult to determine which information from the images contributes the most to the prediction: the distribution of gray-levels, the tumour heterogeneity, the shape of the lesions or the intensities of peritumoural regions. The purpose of this study is to dissociate the different sources of information to improve prediction results. Based on pre-treatment MR images from 103 patients, four types of 3D Volumes Of Interest were defined and arranged in multiple combinations. Combining features extracted from different regions proved to increase prediction performances.

Keywords: Health Informatics - Decision support methods and systems, General and theoretical informatics - Statistical data analysis
Abstract: This study investigates the adoption of innovative Motor and Cognitive Dual-Task (MCDT) based on the combination of increasing motor and cognitive tasks to discern between subjects with Mild Cognitive Impairment (MCI) and Cognitively Normal Adults (CNA). We aim to adopt new MCDT protocols and to compare their performance against the gold standard (a walking based MCDT, called GAIT). 27 older adults have been assessed through a customized wearable system during 4 MCDTs. We developed as many pooled indices (PIs), based on MCDTs perfomance, demographic data, and clinical scores. We use these parameters as regressors in 4 different logistic regression models. The regression models that encompassed features from innovative MCDT overcame the gold standard classification performance. In particular, models based on the heel tapping and the alternate heel-toe tapping reach the best outputs, namely +8% of accuracy if compared to the gold standard (a walking task). The use of logistic regression models based on MCDT PI have been effective in discerning between CNA textit{vs} MCI. Our results suggest that the gold standard MCDT may represents a too demanding exercise to highlight differences between CNA and MCI. It seems that MCDT based on an intermediate level of motor difficulty could represent the sweet spot for the identification of MCI against CNA.

Keywords: Health Informatics - eHealth, Health Informatics - Disease profiling and personalized treatment, General and theoretical informatics - Deep learning and big data to knowledge
Abstract: The study of stress and its implications has been the focus of interest in various fields of science. Many automated/semi-automated stress detection systems based on physiological markers have been gaining enormous popularity and importance in recent years. Such non voluntary physiological features exhibit unique characteristics in terms of reliability, accuracy. Combined with machine learning techniques, they offer a great field of study of stress identification and modelling. In this study, we explore the use of Convolutional Neural Networks (CNN) for stress detection through surface electromyography signals (sEMG) of the trapezius muscle. One of the main advantages of this model is the use of the sEMG signal without computed features contrary to classical machine learning algorithms. The proposed model achieved good results with 73% f1-score for a multi-class classification and 82% in a bi-class classification.

Keywords: Health Informatics - eHealth, Public Health Informatics - Public health management solutions, Health Informatics - Computer-aided decision making
Abstract: Screening for atrial fibrillation (AF) could reduce the incidence of stroke by identifying undiagnosed AF and prompting anticoagulation. However, screening may involve recording many electrocardiograms (ECGs) from each participant, several of which require manual review which is costly and time-consuming. The aim of this study was to investigate whether the number of ECG reviews could be reduced by using a model to prioritise ECGs for review, whilst still accurately diagnosing AF. A multiple logistic regression model was created to estimate the likelihood of an ECG exhibiting AF based on the mean RR-interval and variability in RR-intervals. It was trained on 1,428 manually labelled ECGs from 144 AF screening programme participants, and evaluated using 11,443 ECGs from 1,521 participants. When using the model to order ECGs for review, the number of reviews for AF participants was reduced by 74% since no further reviews are required after an AF ECG is identified; however, it did not impact the number of reviews in non-AF participants (the vast majority of participants), so the overall number of reviews was reduced by 3% with no missed AF diagnoses. When using the model to also exclude ECGs from review, the overall number of reviews was reduced by 28% with no missed AF diagnoses, and by 53% with only 4% of AF diagnoses missed. In conclusion, the workload can be reduced by using a model to prioritise ECGs for review. Ordering ECGs alone only provides only a moderate reduction in workload. The additional use of a threshold to exclude ECGs from review provides a much greater reduction in workload at the expense of some missed AF diagnoses.

Keywords: Health Informatics - Patient tracking, Imaging Informatics - Image analysis, processing and classification, General and theoretical informatics - Algorithms
Abstract: Remote photoplethysmography (PPG) estimates vital signs by measuring changes in the reflected light from the human skin. Compared to traditional PPG techniques, remote PPG enables contactless measurement at a reduced cost. In this paper, we propose a novel method to extract remote PPG signals and heart rate from videos. We propose an algorithm to dynamically track regions of interest (ROIs) and combine the signals from all ROIs based on signal qualities. To maintain a stable frame rate and accuracy, we propose a dynamic down- sampling approach, which makes our system robust to the different video resolutions and user-camera distances. We also propose the strategy of adaptive measurement time to estimate HR, which can achieve comparable accuracy in HR estimation while reducing the average measurement time. To test the accuracy of the proposed system, we have collected data from 30 subjects with facial masks. Experimental results show that the proposed system can achieve 3.0bpm mean absolute error in HR estimation.

Keywords: Sensor Informatics - Multi-sensor data fusion, General and theoretical informatics - Data quality control, Sensor Informatics - Physiological monitoring
Abstract: Numerous applications require accurate estimation of respiratory timings. Respiratory effort (RSP) measurement is a popular approach to accomplish this, especially when the tightness of the sensing belt around the chest can be ensured. In less controlled settings, however, belt looseness and artifacts from movement of the belt on the chest can corrupt the signal. This paper demonstrates that respiration quality indexing and outlier removal can help mitigate these issues, improving estimates of respiration rate (RR), inspiration time (Ti), and expiration time (Te). In a sample of 15 healthy human participants undergoing a protocol of five controlled breathing exercises in four postures each, electrocardiogram (ECG) and RSP signals were collected. RSP signals were processed to extract breath-by-breath estimates of RR, Ti, and Te. These estimates were compared against ground truth spirometry-based estimates using Bland-Altman analysis. We find that incorporating quality indexing and outlier removal prior to feature extraction improves the 95% limits of agreement by 10-40%. We also find that by using ECG-derived respiration (EDR) during periods of RSP artifact, the data removal necessary for accurate respiratory timing estimation is significantly reduced (P < 0.05 for all postures). These findings encourage the use of quality assessment and EDR to enhance the robustness of RR, Ti, and Te estimation from RSP signals. Clinical Relevance— Detecting stimulus-induced or pathological changes in respiratory function can enhance our understanding and monitoring of respiratory health. Quality assessment and the use of EDR help accomplish this by enabling more accurate measurement of respiratory timings.

Keywords: Sensor Informatics - Multi-sensor data fusion, Bioinformatics - Bioinformatics for health monitoring, Sensor Informatics - Sensor-based mHealth applications
Abstract: Abstract— Accurately recognizing health-related conditions from wearable data is crucial for improved healthcare outcomes. To improve the recognition accuracy, various approaches have focused on how to effectively fuse information from multiple sensors. Fusing multiple sensors is a common scenario in many applications, but may not always be feasible in real-world scenarios. For example, although combining bio- signals from multiple sensors (i.e., a chest pad sensor and a wrist wearable sensor) has been proved effective for improved performance, wearing multiple devices might be impractical in the free-living context. To solve the challenges, we propose an effective more to less (M2L) learning framework to improve testing performance with reduced sensors through leveraging the complementary information of multiple modalities during training. More specifically, different sensors may carry different but complementary information, and our model is designed to enforce collaborations among different modalities, where positive knowledge transfer is encouraged and negative knowledge transfer is suppressed, so that better representation is learned for individual modalities. Our experimental results show that our framework achieves comparable performance when compared with the full modalities. Our code and results will be available at https://github.com/comp-well-org/More2Less.git.

Keywords: Cardiovascular, respiratory, and sleep devices - Wearables
Abstract: Wearable bioimpedance is a technique proposed to estimate breathing parameters such as respiratory rate (RR). However, its potential application lies in clinical investigation of daily-life activities like walking. This study evaluated the effect of the walking interference on the estimation of breathing parameters. 50 chronic obstructive pulmonary disease patients performed static and active measurements during thoracic bioimpedance acquisition. The static measurements included respiratory airflow for reference. The active measurements were used to estimate the walking interference from bioimpedance, and the obtained signals were added to static measurements for comparison with the reference. Afterward, we applied four different preprocessing methods to remove this walking interference and the resulting signals were used to detect the respiratory cycles and estimate breathing parameters (inspiratory time, expiratory time, duty cycle, and RR). The methods performed differently in terms of accuracy and mean average percentage error (MAPE), showing the need for specific preprocessing for active measurements. Furthermore, the MAPE values in the RR estimation were close to 3 % indicating that breathing parameters can be accurately estimated during walking. Accordingly, the present study reinforces the applicability of wearable bioimpedance for respiratory monitoring.

Keywords: Pulmonary and critical care - Bioengineering applications in Intensive care, Pulmonary and critical care - Pulmonary disease, Cardiovascular and respiratory system modeling - Gas exchange models
Abstract: We present new results validating the capability of a high-fidelity computational simulator to accurately predict the responses of individual patients with acute respiratory distress syndrome to changes in mechanical ventilator settings. 26 pairs of data-points comprising arterial blood gasses collected before and after changes in inspiratory pressure, PEEP, FiO2, and I:E ratio from six mechanically ventilated patients were used for this study. Parallelized global optimization algorithms running on a high-performance computing cluster were used to match the simulator to each initial data point. Mean absolute percentage errors between the simulator predicted values of PaO2 and PaCO2 and the patient data after changing ventilator parameters were 10.3% and 12.6%, respectively. Decreasing the complexity of the simulator by reducing the number of independent alveolar compartments reduced the accuracy of its predictions.

Keywords: Pulmonary and critical care - Ventilatory Assist Devices, Pulmonary and critical care - Bioengineering applications in Intensive care, Respiratory transport, mechanics and control - Periodic breathing
Abstract: The magnitude of inspiratory effort relief within the first 2 hours of non-invasive ventilation for hypoxic respiratory failure was shown in a recent exploratory clinical study to be an early and accurate predictor of outcome at 24 hours. We simulated the application of non-invasive ventilation to three patients whose physiological and clinical characteristics match the data in that study. Reductions in inspiratory effort corresponding to reductions of esophageal pressure swing greater than 10 cmH2O more than halved the values of total lung stress, driving pressure, power and transpulmonary pressure swing. In the absence of significant reductions in inspiratory pressure, multiple indicators of lung injury increased after application of non-invasive ventilation.

Keywords: Pulmonary and critical care - Pulmonary function testing & instrumentation, Cardiovascular, respiratory, and sleep devices - Sensors, Cardiovascular, respiratory, and sleep devices - Monitors
Abstract: There is growing research showing the importance of measuring esophageal pressure as a surrogate for pleural pressure for patients on mechanical ventilators. The most common measurement method uses a balloon catheter whose accuracy can vary based on patient anatomy, balloon position, balloon inflation, and the presence of other tubes in the esophagus. The authors present the development and initial testing results of a new combination catheter, utilizing fiberoptic pressure sensing to provide more accurate esophageal pressure measurements, and allows for the incorporation of a feeding tube and temperature sensor.

Keywords: Cardiovascular and respiratory signal processing - Cardiovascular signal processing
Abstract: In this study, we proposed an automatic detector for obstructive apnea episodes using only ECG-based time-series from a single-ECG channel. Several obstructive apnea episodes were provoked for different separated sequences of 15 minutes in anesthetized Sprague-Dawley rats. In this recurrent obstructive sleep apnea (OSA) model, each episode lasted 15 s, while the number of total events per sequence was randomly selected. The beat-to-beat interval (RR) and the R-wave amplitude (Ra) time-series were extracted and pre-processed for each sequence and used to train Dynamic Bayesian Networks (DBN) with different lags. An optimal trade-off between the lag (L) and RMSE values was considered to select the best model to be used when detecting apnea episodes. The selected models were then used to estimate the occurrence probability of apnea episodes, p(At), by using a filtering approach. Finally, the time series of the estimated probabilities were post-processed using non-overlapped 15-s epochs, to determine whether they are classified as apneic or non-apneic segments. Results showed that those lagged models with orders greater than 5, presented suitable RMSE values and become more sensitive as the order increased. A detection threshold of 0.2 seems to provide the best apnea detection performance overall, with Acc=0.81, Se=0.83, and Sp=0.79, using two ECG parameters and L = 10. Clinical relevance— Dynamic Bayesian Networks represent a powerful tool to develop personalized models for apnea detection and diagnosis in OSA patients.

Keywords: Pulmonary and critical care - Pulmonary function testing & instrumentation, Cardiovascular and respiratory signal processing - Time-frequency, time-scale analysis of respiratory variability
Abstract: Electrical impedance tomography (EIT) is a bio-medical imaging modality that has several clinical applications namely for human lungs. Yet, its relationship with gold standard lung diagnostic tools including spirometry is not available. In this study, simultaneous EIT and spirometry measurements were collected for 14 healthy subjects who performed forced breathing paradigms of different efforts simulating a wide range of spirometry indicators. It is demonstrated that EIT can predict standard spirometry indicators over a wide dynamic range, with a potential sensitivity and specificity of 98% and 100%, respectively, in detecting obstructive patterns. It is also shown that EIT can provide a regional mapping of the spirometry indicator which are shown to be consistent with their corresponding global indicators. Overall, EIT can predict spirometry indicators and can assess regional lung health through parametric mapping.

Keywords: Teaching design, Novel approaches to BME education, Career development in BME
Abstract: The Undergraduate Program in Biomedical Engineering ITB, Indonesia, introduce the Health Technology Assessment and Management as an elective course in 2021. This course is implemented to support the World Health Assembly that urges the member states to establish national strategies in health technology assessment and management, particularly medical devices. Furthermore, it is designed to give biomedical engineering students a broader insight into their career opportunities. Therefore, this course is delivered by the practitioner and guided by the main lecturer. The course syllabus is developed from the WHO Medical Devices Technical Series and European Network for Health Technology Assessment. It tries to implement HTA Core Model for Rapid Relative Effectiveness Assessments. A questionnaire is used to measure the students' perception of the course implementation. Moreover, it is used to obtain the students' comments and feedback. The course that is delivered by the practitioner not only gives the course content but also the context. After attending the course, students have a broader insight into the career opportunities as biomedical engineers in Indonesia.

Keywords: Teaching design, Instruction and learning, Novel approaches to BME education
Abstract: The need for a training modality for tooth extraction procedures is increasing, as dental students do not feel properly trained. In this study, a prototype of a training setup is designed, in which extraction procedures can be performed on jaw models and cadaveric jaws. The prototype was designed in a way that it can give real-time feedback on the applied forces in all three dimensions (buccal/lingual, mesial/distal, and apical/coronal), torques, and angular velocity. To evaluate the prototype, a series of experimental extractions on epoxy models, conserved jaws, and fresh frozen jaws were performed. Extraction duration (s), angular velocity (degrees/s), average force (N), average torque (Nm), linear impulse (Ns), and angular impulse (Nms) were shown in real-time to the user and used to evaluate the prototype. In total, 342 (92.9%) successful extractions were performed using the prototype (n= 113 epoxy factory-made, n=187 epoxy re-used, n=17 conserved, n=25 fresh frozen). No significant differences were found between the conserved and the fresh frozen jaws. The fresh frozen extraction duration, linear impulse, and angular impulse differed significantly from the corresponding values obtained for the epoxy models. Extractions were successfully performed, and the applied forces, torques, and angular velocity were recorded and shown as real-time feedback using the prototype of the dental extraction trainer. The feedback of the prototype is considered reliable.

Keywords: Career development in BME, Novel approaches to BME education, Instruction and learning
Abstract: The European Medical Device and In-Vitro Diagnostic Medical Device industry faces currently a highly challenging situation, as applied regulations changed throughout the last year significantly. To cope with this situation a novel continuous education program for employees from these sectors, aiming to provide knowledge and hands-on experience in clinical, regulatory, and quality affairs. Since 2020 two classes were successfully completed at the University of Applied Sciences and Arts Northwestern Switzerland. Within this paper the concept and content are described, the participants population highlighted, and the benefit for the participant and their employer elaborated.

Keywords: Biomedical engineering curricula, Teaching design, BME undergraduate research
Abstract: Innovation and development in medical devices are of great significance for overcoming the problems such as the increasing costs of healthcare and the widening societal inequity in medical technologies. This paper presents the design and outcomes of a general education course in medical devices innovation offered for an eight-week quarter each year since Spring 2018 at Zhejiang University. The course consists of two modules, lectures and team project, both of which are well designed based on the entire innovation process spanning needs finding, concept generation, prototyping, and strategy development. A professional teaching team with eight experts from various disciplines and institutes has been established as well. Since its inception, 296 students from 34 majors have participated and 71 original projects have been proposed. The results of self-assessment questionnaires showed that the course had equipped students with broader fundamentals and specialized knowledge, and stronger skills in innovation and teamwork, which provide a solid foundation for students' future innovation practice.

Keywords: Novel approaches to BME education, BME and global health, Instruction and learning
Abstract: Abstract— Healthcare Innovation ideas originating from biomedical engineering departments are rarely based on a deep understanding of a problem, but are often based on coming up with an engineering solution that does not meet an Unmet Clinical Need, is too complicated, bulky, costly, and does not consider global developments. For an impactful innovation design it is essential however to properly understand the clinical issues, forward project the effect of exponential technologies and other global developments. Health and healthcare are in need of disruptive ideas for preventive, predictive, personalised solutions that engage the individuals to pave the way towards real healthcare. We have adapted a novel meta-methodology for dedicated use with health related applications and have used it validating start-up ideas and also during a semester long lecture/seminar classroom setup with amazing results.

Clinical Relevance— This novel health dedicated meta-methodology is dependent on interdisciplinary team and innovation work and heavily relies on a good understanding of the current clinical processes and needs as well as on a future projection of global health delivery developments. The clinical perspective is essential and meaning- and impactful innovation can only be developed validating desirability feasibility, and viability, which needs clinical-, engineering/technical-, as well as economic expertise.

Keywords: Point of care - Detection and monitoring, Medical technology - Entrepreneurship and commercialization, Medical technology - Innovation
Abstract: Intravenous (IV) infiltration is a common problem associated with IV infusion therapy in clinical practice. A multitude of factors can cause the leakage of IV fluids into the surrounding tissues, resulting in symptoms ranging from temporary swelling to permanent tissue damage. Severe infiltration outcomes can be avoided or minimized if the patient’s care provider is alerted of the infiltration at its earliest onset. However, there is a lack of real-time, continuous infiltration monitoring solutions, especially those suited for clinical use for critically ill patients. Our design of the sensor-integrated ATTENTIV catheter allows direct detection of catheter dislodgement, a root cause of IV infiltration. We verify two detection methods: blood-tissue differentiation with a support vector machine and signal peak identification with a thresholding algorithm. We present promising preliminary testing results on biological and phantom models that utilize bioimpedance as the sensing modality.

Keywords: Medical technology - Product development process, Medical technology - Design and development, Medical technology - Innovation
Abstract: This paper introduces design modifications to our MR-Conditional, 2-degree-of-freedom (DOF), remotely- actuated needle driver for MRI-guided spinal injections. The new needle driver should better meet cleaning and sterilization guidelines needed for regulatory approval, preserve the sterile field during intraoperative needle attachment, and offer better ergonomics and intuitiveness when handling the device. Dynamic and static force and torque required to properly install the needle driver onto our 4-DOF robot base are analyzed, which provide insight into the risks of intraoperative tool attachment in the setting of robot-assisted spinal injections under MRI guidance.

Keywords: Medical technology - Innovation, Medical technology - Design and development
Abstract: Loosening of pedicle screws after spinal fusion surgery can prevent the desired fusion between vertebrae and may be reason for revision surgery. Especially in osteoporotic bone, toggling of pedicle screws is a common problem which compromises the fixation strength of these screws and can lead to loosening or axial pull-out of the screw. In this study we explore the use of an in-pedicle expandable anchor that shapes to the pedicle to increase the toggling resistance of the anchor by increasing the contact area between the anchor and the dense cortical bone of the pedicle. A scaled up, two dimensional prototype was designed. The prototype consists of a bolt and ten stainless steel wedges that expand by tensioning the bolt. During the expansion, the wedges are required to compress the cancellous bone. Based on the first preliminary experiment, it was found that the expansion of the wedges resulted in successful compression of 5 PCF cancellous bone phantom (Sawbones). This preliminary study shows that an expandable in-pedicle anchor could be a feasible option to increase the toggling resistance of spinal bone anchors, especially in osteoporotic bone. Clinical Relevance— Toggling of pedicle screws is a major cause of screw loosening. In this preliminary study, the use of an in-pedicle expandable anchor to increase the toggling resistance of spinal bone anchors is explored.

Keywords: Medical technology - Innovation, Medical technology - Product development process, Point of care - Home-based applications
Abstract: Wearable medical technologies (wearables) are a type of digital health technology that have the potential to transform the delivery of healthcare by offering a personalized, predictive, and preventative form of patient care. Unfortunately, the adoption of these technologies remains low, indicating an issue with the innovation ecosystem. Existing literature on innovation ecosystems fails to address wearables and the trade-offs between profitability and health outcomes, hindering discussion on how to promote the widespread adoption of wearables. This paper reviews and synthesizes literature from the business community and healthcare technology community on innovation ecosystems to contribute towards developing a more robust model of the innovation ecosystem for developing wearables. Future work aims to enrich this conceptualization with interviews of key actors in the innovation ecosystem.

Keywords: Empowering individual healthcare decisions through technology, Preventive medicine, Medical technology - Clinical testing/clinical trials
Abstract: Low-frequency Fahræus–Lindqvist-driven oscillations in the small vessels are crucial because oscillations in small vessels support nutrient supply. Understanding of this is critical in type 2 diabetes mellitus (T2DM) to develop therapeutic measures in order to prevent Alzheimer's Disease Related Dementias. Indeed, vascular factors are known to contribute to cerebrovascular disease as well as mild cognitive impairment and dementia, which are predicted to affect 152 million people by 2050 (Alzheimer's Disease International London, UK, 2019). In this clinical study, we performed functional near-infrared spectroscopy (fNIRS) of the forehead to investigate the effect of the Mini-Cog with three-item recall test on the prefrontal cortex (PFC) activation and the relative oscillatory power in the 0.01–0.02-Hz (Fahræus–Lindqvist effect) and 0.021–0.052 Hz (smooth muscle autonomic innervation) frequency bands in elderly (60 years and older) T2DM and age-matched controls. We found a significant (p<0.01) difference in the PFC activation between elderly subjects with T2DM and age-matched elderly controls. Moreover, power spectral density (PSD) analysis revealed a significantly lower relative power in 0.021–0.052 Hz (smooth muscle autonomic innervation) frequency band in elderly subjects with T2DM during the Mini-Cog three-item recall test. Furthermore, a drop in the oscillatory power in the 0.01–0.02 Hz frequency band during Mini-Cog three-item recall test was found more pronounced in the elderly subjects with T2DM. Therefore, our study highlighted portable brain imaging to capture cerebrovascular reactivity to cognitive load that may provide a biomarker of cerebrovascular dysfunction in T2DM.

Keywords: Brain-computer/machine interface, Neural signals - Machine learning & Classification, Neural signals - Nonlinear analysis
Abstract: Kernel temporal differences (KTD) (𝛌) algorithm integrated in Q-learning (Q-KTD) has shown its applicability and feasibility for reinforcement learning brain machine interfaces (RLBMIs). RLBMI with its unique learning strategy based on trial-error allows continuous learning and adaptation in BMIs. Q-KTD has shown good performance in both open and closed-loop experiments for finding a proper mapping from neural intention to control commands of an external device. However, previous studies have been limited to intracortical BMIs where monkey’s firing rates from primary motor cortex were used as inputs to the neural decoder. This study provides the first attempt to investigate Q-KTD algorithm’s applicability in EEG-based RLBMIs. Two different publicly available EEG data sets are considered, we refer to them as Data set A and Data set B. EEG motor imagery tasks are integrated in a single step center-out reaching task, and we observe the open-loop RLBMI experiments reach 100% average success rates after sufficient learning experience. Data set A converges after approximately 20 epochs for raw features and Data set B shows convergence after approximately 40 epochs for both raw and Fourier transform features. Although there still exist challenges to overcome in EEG-based RLBMI using Q-KTD, including increasing the learning speed, and optimization of a continuously growing number of kernel units, the results encourage further investigation of Q-KTD in closed-loop RLBMIs using EEG.

Keywords: Neurological disorders, Brain functional imaging
Abstract: Background: Our previous studies have proved that preclinical Alzheimer's disease (AD) which including subjective cognitive decline (SCD) stage, can be distinguished from normal control (NC) by glucose-oxygen metabolism coupling at the voxel level, but whether the coupling at the network level worked has not been studied. Therefore, this study aimed to explore the coupling relationship between brain glucose metabolic connectivity network and oxygen functional connectivity network, and whether its feasibility as a biomarker to discriminate SCD from healthy control (HC). Methods: Resting-state functional magnetic resonance imaging (rs-fMRI) and glucose positron emission tomography (PET) based on hybrid PET/MRI scans were used to investigate metabolism-oxygen metabolism coupling in 56 SCD individuals and 54 HCs. Network coupling features were selected by logistic regression-recursive feature elimination (LR-RFE), and then a linear support vector machine (SVM) was used to distinguish SCD and HC by using 5-fold cross-validation. Results: The classification average accuracy of network coupling had reached 76.36% with a standard deviation of 9.85% (with a sensitivity of 77.82% ± 15.13% and a specificity of 75.30% ± 15.15%). After receiver operating characteristic (ROC) analysis, the average area under curve (AUC) of network coupling was 0.788 (95% confidence interval [CI] = [0.653-0.983]). Conclusion: This study provided a new perspective for exploring network coupling. The proposed classification method highlighted the potential clinical application by combing glucose-oxygen metabolism coupling and machine learning in identifying SCD.

Keywords: Brain functional imaging - EEG, Neural signal processing, Brain-computer/machine interface
Abstract: Olfactory perception is shaped by dynamic interactions among networks of widely distributed brain regions involved in several neurocognitive processes. However, the neural mechanisms that enable effective coordination and integrative processing across these brain region, which have different functions and operating characteristics, are not yet fully understood. In this study we use electroencephalography (EEG) signals and a multilayer network formalism to model cross-frequency coupling across the brain and identify brain regions that operate as connecting hubs, thus facilitating integrative function. To this goal we investigate α − γ coupling and θ − γ coupling during exposure to olfactory stimuli of different pleasantness levels. We found that a wider distributed network of hubs emerges in the higher pleasantness condition and that significant differences in the hubs connectivity are located in the middle frontal and central regions. Our results indicate the consistent functional role that γ band activity plays in information integration in olfactory perception.

Keywords: Human performance - Modelling and prediction
Abstract: Electroencephalography (EEG) signals can effectively measure the level of human decision confidence. However, it is difficult to acquire EEG signals in practice due to the expensive cost and complex operation, while eye movement signals are much easier to acquire and process. To tackle this problem, we propose a Cross-modality deep learning method based on Deep Canoncial Correlation Analysis (CDCCA) to transform each modality separately and coordinate different modalities into a hyperspace by using specific canonical correlation analysis constraints. In our proposed method, only eye movement signals are used as inputs in the test phase and the knowledge from EEG signals is learned in the training stage. Experimental results on two human decision confidence datasets demonstrate that our proposed method achieves advanced performance compared with the existing single-modal approaches trained and tested on eye movement signals and maintains a competitive accuracy in comparison with multimodal models.

Keywords: Brain-computer/machine interface, Neural signal processing
Abstract: Reinforcement learning (RL)-based brain-machine interfaces (BMIs) learn the mapping from neural signals to subjects’ intention using a reward signal. External rewards (water or food) or internal rewards extracted from neural activity are leveraged to update the parameters of decoders in the existing RL-based BMI framework. However, for complex tasks, the design of external reward could be difficult, which may not fully reflect the subject’s own evaluation internally. It is important to obtain an internal reward model from neural activity to access subject’s internal evaluation when the subject is performing the task through trial and error. In this paper, we propose to use an inverse reinforcement learning (IRL) method to estimate the internal reward function interpreted from the brain to assist the update of the decoders. Specifically, the inverse Q-learning (IQL) algorithm is applied to extract internal reward information from real data collected from medial prefrontal cortex (mPFC) when a rat was learning a two-lever-press discrimination task. Such an internal reward information is validated by checking whether it can guide the training of the RL decoder to complete movement task. Compared with the RL decoder trained with the external reward, our approach achieves a similar decoding performance. This preliminary result validates the effectiveness of using IRL to obtain the internal reward model. It reveals the potential of estimating internal reward model to improve the design of autonomous learning BMIs.

Keywords: Motor learning, neural control, and neuromuscular systems, Neuromuscular systems - Computational modeling, Brain physiology and modeling - Neural dynamics and computation
Abstract: Goal-driven networks trained to perform a task analogous to that performed by biological neural populations are being increasingly utilized as insightful computational models of motor control. The resulting dynamics of the trained networks are then analyzed to uncover the neural strategies employed by the motor cortex to produce movements. However, these networks do not take into account the role of sensory feedback in producing movement, nor do they consider the complex biophysical underpinnings of the underlying musculoskeletal system. Moreover, these models can not be used in context of predictive neuromechanical simulations for hypothesis generation and prediction of neural strategies during novel movements. In this research, we adapt state-of-the-art deep reinforcement learning (DRL) algorithms to train a controller to drive a developed anatomically accurate monkey arm model to track experimentally recorded kinematics. We validate that the trained controller mimics biologically observed neural strategies to produce movement. The trained controller generalizes well to unobserved conditions as well as to perturbation analyses. The recorded firing rates of motor cortex neurons can be predicted from the controller activity with high accuracy even on unseen conditions. Finally, we validate that the trained controller outperforms existing goal-driven and representational models of motor cortex in single neuron decoding accuracy, thus showing the utility of the complex underpinnings of anatomically accurate models in shaping motor cortex neural activity during limb movements. The learned controller can be used for hypothesis generation and prediction of neural strategies during novel movements and unobserved conditions.

Keywords: Point of care - Respiratory monitoring, Point of care - Home-based applications, Point of care - Clinical use and acceptance
Abstract: The use of smartphones in clinical practice is referred to as mobile health (mHealth). This has attracted great interest in both academia and industry because of its potential to augment healthcare. In this study, we developed an mHealth app for the non-contact measurement of chest-wall movements using the iPhone ’s built-in depth sensor, thereby enabling a pulmonary self-monitoring function for personal use. The depth sensor provides depth values for each pixel and 2D mapping of the chest-wall movements. To extract respiratory signals from the right and left thoracic regions and abdomen, a 2D-depth image-segmentation method was implemented. The method was based on the anatomy and physiology of chest-wall movements, assuming differences in the anterior displacement in the thoracic and abdominal regions. It was observed that the differences were significant (p < 0.05) in the segmented regions of interest (ROIs) of the right and left thoracic region and abdomen. Respiratory signals extracted from each ROI were compared with the contact bio-impedance signals, which were highly correlated (r = 0.94).

Keywords: Point of care - Home-based applications, Medical technology - Design and development, Personalized medicine
Abstract: Brain-computer interfaces (BCIs) are emerging as a new solution for individuals with severe disability to interact with the world. However, BCI technologies have yet to reach end-users in their daily lives due to significant translational gaps. To address these gaps, we applied user-centered design principles to establish a home BCI program for children with quadriplegic cerebral palsy. This work describes the technical development of the software we designed to facilitate BCI use at home. Children and their families were involved at each design stage to evaluate and provide feedback. Since deployment, seven families have successfully used the system independently at home and continue to use BCI at home to further enable participation and independence for their children. Clinical relevance: The design and successful implementation of user-centered software for home use will both inform on the feasibility of BCI as a long-term access solution for children with neurological disabilities as well as decrease barriers of accessibility and availability of BCI technologies for end-users.

Keywords: Point of care - Detection and monitoring, Point of care - Home-based applications, Empowering individual healthcare decisions through technology
Abstract: In-bed behavior monitoring is commonly needed for bed-bound patient and has long been confined to wearable devices or expensive pressure mapping systems. Meanwhile, vision-based human pose and posture tracking while experiencing a lot of attention/success in the computer vision field has been hindered in terms of usability for in-bed cases, due to huge privacy concerns surrounding this topic. Moreover, the inference models for mainstream pose and posture estimation often require excessive computing resources, impeding their implementation on edge devices. In this paper, we introduce a privacy-preserving in-bed pose and posture tracking system running entirely on an edge device with added functionality to detect stable motion as well as setting user-specific alerts for given poses. We evaluated the estimation accuracy of our system on a series of retrospective infrared (LWIR) images as well as samples from a real-world test environment. Our test results reached over 93.6% estimation accuracy for in-bed poses and achieved over 95.9% accuracy in estimating three in-bed posture categories.

Keywords: Point of care - Evaluation and validation, Point of care - Clinical use and acceptance, Point of care - Heart rate monitoring
Abstract: Wearable heart rate monitors offer a cost-effective way of non-invasive, long-term monitoring of cardiac health. Validation of wearable technologies in an older populations is essential for evaluating their effectiveness during deployment in healthcare settings. To this end, we evaluated the validity of heart rate measures from a wearable device, Empatica E4, and compared them to the electrocardiography (ECG). We collected E4 data simultaneously with ECG in thirty-five older men and women during an overnight sleep recording in the laboratory. We propose a robust approach to resolve the missing inter-beat interval (IBI) data and improve the quality of E4 derived measures. We also evaluated the concordance of heart rate (HR) and heart rate variability (HRV) measures with ECG. The results demonstrate that the automatic E4 heart rate measures capture long-term HRV whilst the short-term metrics are affected by missing IBIs. Our approach provides an effective way to resolve the missing IBI issue of E4 and extracts reliable heart rate measures that are concordant with ECG.

Keywords: Point of care - Diagnostics, Point of care - Detection and monitoring, Empowering individual healthcare decisions through technology
Abstract: A dipstick urinalysis test is performed by immersing a reagent strip in the urine specimen and then comparing the resulting reagent pad colors with a reference key. The color assessment of the reagent strip can be performed manually or by using a urine analyzer. However, the manual procedure is prone to subjective inaccuracies in varying ambient illumination and urine analyzer equipment is expensive. This paper presents a smartphone-based machine-learning approach to accurately determine the reagent pad colors for automated assessment. We start with a unique calibration chart and use multivariate linear regression to map the captured color values to their true equivalents. This accounts for the camera-induced distortions and ambient illumination factors. Subsequently, the color comparison is performed using the least Euclidean distance to match the calibrated color of each reagent pad with the reference key. The results from an experimental study, using five different smartphone cameras and three common illumination settings, indicate a high degree of accuracy in color assessment for synthetic dipsticks. The proposed smartphone-based method is an easy-to-perform, time-efficient, and cost-effective solution for an automated urinalysis and could be used as an alternative to manual reading or benchtop urine analyzers.

Keywords: Point of care - Home-based applications
Abstract: Chronic kidney disease (CKD) is an escalating global health concern, and non-invasive means for early CKD detection is eagerly awaited. Here, we explore the potential of using home-based frequency-difference electrical impedance tomography (fdEIT) to evaluate CKD based on bio-conductivity characteristics. We first verified the feasibility of using portable EIT capturing bio-conductivity in fresh pig kidneys ex vivo. We further performed bio-conductivity measurement in vivo paired with standard eGFR measurements on CKD patients by EIT and traditional blood test, respectively. Our results showed a significant correlation between renal bio-conductivity changes captured by fdEIT and standard eGFR scores. These results hold promise to be developed into a non-invasive and portable device for early CKD detection and longitudinal CKD treatment monitoring in clinical, community and home-based settings.

Keywords: Data mining and big data methods - Biosignal classification, Data mining and big data methods - Pattern recognition, Data mining and big data methods - Machine learning and deep learning methods
Abstract: Photoplethysmography (PPG) signal comprises physiological information related to cardiorespiratory health. High-quality PPG signals are necessary to extract cardiores piratory information accurately. Motion artifacts can easily corrupt PPG signals due to human locomotion, leading to noise enriched, poor quality signals. Several rule-based and Machine-Learning (ML) - based approaches for PPG signal quality estimation are available, but those algorithms’ efficacy is questionable. So, the authors propose a lightweight CNN architecture for signal quality assessment by employing a novel Quantum Pattern Recognition (QPR) technique. The proposed algorithm is validated on manually annotated data obtained from the University of Queensland database. A total of 28366, 5s signal segments are preprocessed and transformed into image files of 20 × 500 pixels for input to the 2D CNN architecture. The developed model classifies the PPG signal as ’good’ and ’bad’ with an accuracy of 98.3% with 99.3% sensitivity, 94.5% specificity and 98.9% F1-score. The experimental analysis concludes that slim architecture and novel Spatio-temporal pattern recognition technique improve the system’s performance. The proposed approach is suitable for a resource-constrained wearable implementation.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification
Abstract: Photoplethysmography (PPG) is a non-invasive technique used in wearable devices to collect various vital signs, including heart rate and heart rate variability. The signal is highly susceptible to motion artifacts, which is inevitable in health monitoring and may lead to inaccurate decision-making. Studies in the literature proposed time series analysis, signal decomposition, and machine learning methods to reconstruct PPG signals or reduce noise. However, they are limited to short-term noisy signals or to noise caused by certain physical activities. In this paper, we propose a deep convolutional generative adversarial network (GAN) method to reconstruct distorted PPG signals. Our method exploits the temporal information extracted from the corrupted signal and preceding data to perform PPG reconstruction. The model is trained and tested using data collected by smartwatches in a home-based health monitoring application. We evaluate the proposed GAN method in comparison to three state-of-the-art PPG reconstruction methods. The evaluation includes noisy PPG signals with different durations and SNR values. The proposed method outperforms the other methods by obtaining the least error rates. The results indicate that the proposed method is effective for improving PPG signal quality to produce reliable heart rate and heart rate variability.

Keywords: Data mining and big data methods - Machine learning and deep learning methods, Neural networks and support vector machines in biosignal processing and classification
Abstract: Ambulatory respiration signal extraction system is required to maintain continuous surveillance of a patient with respiratory deficiency. The capnograph signal has received a lot of attention in recent years as a valuable indicator of respiratory conditions. However, the typical capnograph signal extraction method is quite expensive and also unpleasant to the patient due to the involvement of a nasal cannula. With the advent of wearable sensor technology, there has been significant research on the use of photoplethysmogram (PPG) signals as a less expensive alternative to extract respiratory information. In this paper, we propose CapNet, a novel deep learning-based framework which takes the regular PPG signal as input, and estimates the capnograph signal as output. Training, validation and testing of the proposed networks in CapNet is done using the IEEE TMBE Respiratory Rate Benchmark dataset by utilizing reference capnograph respiration signals. With a lower MSE and higher cross-correlation values, CapNet outperforms two traditional signal processing algorithms and another recently proposed deep neural network, RespNet. The proposed framework expectantly can be implementable and feasible for constant supervising of patients undergoing respiratory ailments.

Keywords: Data mining and big data methods - Pattern recognition, Signal pattern classification, Nonlinear dynamic analysis - Biomedical signals
Abstract: The study of plethysmography time series is crucial to better understand the breathing behavior of mice, in particular the influence of neurotoxins on the respiratory system. Current approaches rely on a few respiratory descriptors computed on individual breathing cycles that fail to account for the variety of breathing habits and their evolution with time. In this paper we introduce a new procedure for the automatic analysis of plethysmography signals. Our method relies on a new and robust segmentation of respiratory cycles and a DTW-based clustering algorithm to extract the most typical respiratory cycles (called reference sequences). We can then create a symbolic representation of any new recording by matching respiratory cycles to their closest reference sequence. This new representation is a visual and quantitative tool to assess the breathing behavior of mice and its evolution with time. Our method is applied to plethysmography signals collected on mice with two different genotypes and exposed to a neurotoxin.

Clinical relevance: This article proposes a novel approach for studying plethysmography data. Our algorithm is able to accurately extract meaningful respiratory cycles and associated ventilation patterns descriptors such as tidal volume and inhalation/exhalation duration. In addition, thanks to the associated symbolic representation of signals, the temporal evolution of respiration is easily quantified. This opens a new research path to study the often slowly evolving and subtle influence of neurotoxins on the respiratory system.

Keywords: Signal pattern classification, Data mining and big data methods - Biosignal classification
Abstract: Circadian rhythms in blood pressure (BP) may in some cases be indicative of an increased risk of adverse cardiovascular events. However, current methods for assessing these rhythms can be disruptive to sleep, work, and daily activities. Features of the photoplethysmogram (PPG), which can be non-invasively and unobtrusively recorded, have been suggested as surrogate measures of BP. This work investigates the presence of a circadian rhythm in these features and evaluates their potential to classify nocturnal BP patterns. 742 patients who were discharged home from the ICU were selected from the MIMIC-III database. Our results show that a number of PPG features exhibit a clear and observable circadian rhythm. Of the 19 features evaluated, the circadian rhythms of 5 features outperformed heart rate (HR) in terms of correlation with the circadian rhythm of SBP (|r| > 0.81). We also present evidence that a metric combining the PPG features significantly improves BP phenotype classification accuracy.

Keywords: Time-frequency and time-scale analysis - Time-frequency analysis, Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: The importance of accurate and continuous heart rate monitoring during workout cannot be overestimated. Supporting heart rate in the desired range, you strengthen the heart muscle and the overall fitness level. Therefore, the precise heart rate measurements are important at each stage: before workout, during exercises and after completion. One of the most important problems for precise heart rate measurement during high intensive exercises with the help of wearable devices is the movement artifacts. Even the smallest movement of the muscles, the turn of the wrist, the movement of the fingers or even the displacement of the fitness tracker per millimeter - all this very much distorts the useful signal. To solve the problem of motion artifacts, both digital signal processing approaches are used, as well as deep learning methods. In this paper, we offer a new method of processing the signal of wearable devices during workout, in particular to solve the motion artifacts problem, using deep neural networks. A distinctive feature of the model is the use of higher signal harmonics to determine the shape and type of signal. In particular, a method is proposed for classifying the signal spectrogram to noise, movement and useful component. During cross-validation on an available datasets, we compared the effectiveness of the proposed approach for spectrogram classification and received an improvement of averaged ROC AUC (area under the receiver operating characteristic curve) and F1 Score by 5% by using higher harmonics.

Keywords: Neural networks and support vector machines in biosignal processing and classification
Abstract: Abstract — A growing area of mental health research pertains to how an individual's degree of depression might be automatically assessed through analyzing multimodal-based objective markers. However, when combined with machine learning, this research can be challenging due to the existence of unaligned multimodal sequences and the limited amount of annotated training data. In this paper, a novel cross-modal framework for automatic depression severity assessment is proposed. The low-level descriptions (LLDs) from multiple clues (such as text, audio and video) are extracted, after which multimodal fusion via cross-modal attention mechanism is utilized to facilitate the learning of more accurate feature representations. For the features extracted from each modality, the cross-modal attention mechanism is utilized to continuously update the input sequence of the target mode, until the score of the patient's health questionnaire (PHQ-8) can finally be obtained. Moreover, Self-Attention Generative Adversarial Networks (SAGAN) is employed to increase the amount of training data available for depression severity analysis. Experimental results on the depression sub-challenge dataset of the Audio/Visual Emotion Challenge (AVEC 2017 and AVEC 2019) demonstrate the effectiveness of our proposed method.

Keywords: Neural networks and support vector machines in biosignal processing and classification, Physiological systems modeling - Signal processing in physiological systems, Time-frequency and time-scale analysis - Time-frequency analysis
Abstract: This paper presents a speech-based system for autism severity estimation combined with automatic speaker diarization. Speaker diarization was performed by two different methods. The first used acoustic features, which included Mel-Frequency Cepstral Coefficients (MFCC) and pitch, and the second used x-vectors – embeddings extracted from Deep Neural Networks (DNN). The speaker diarization was trained using a Fully Connected Deep Neural Network (FCDNN) in both methods. We then trained a Convolutional Neural Network (CNN) to estimate the severity of autism based on 48 acoustic and prosodic features of speech. One hundred thirty-two young children were recorded in the Autism Diagnostic Observation Schedule (ADOS) examination room, using a distant microphone. Between the two diarization methods, the MFCC and Pitch achieved a better Diarization Error Rate (DER) of 26.91%. Using this diarization method, the severity estimation system achieved a correlation of 0.606 (Pearson) between the predicted and the actual autism severity scores (i.e., ADOS scores).

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: We suggested a unified system with core components of data augmentation, ImageNet-pretrained ResNet-50, cost-sensitive loss, deep ensemble learning, and uncertainty estimation to quickly and consistently detect COVID-19 using acoustic evidence. To increase the model's capacity to identify a minority class, data augmentation and cost-sensitive loss are incorporated (infected samples). In the COVID-19 detection challenge, ImageNet-pretrained ResNet-50 has been found to be effective. The unified framework also integrates deep ensemble learning and uncertainty estimation to integrate predictions from various base classifiers for generalisation and reliability. We ran a series of tests using the DiCOVA2021 challenge dataset to assess the efficacy of our proposed method, and the results show that our method has an AUC-ROC of 85.43 percent, making it a promising method for COVID-19 detection. The unified framework also demonstrates that audio may be used to quickly diagnose different respiratory disorders.

Keywords: Signal pattern classification, Neural networks and support vector machines in biosignal processing and classification
Abstract: This paper presents a deep learning framework for detecting COVID-19 positive subjects from their cough sounds. In particular, the proposed approach comprises two main steps. In the first step, we generate a feature representing the cough sound by combining an embedding extracted from a pre-trained model and handcrafted features extracted from draw audio recording, referred to as the front-end feature extraction. Then, the combined features are fed into different back-end classification models for detecting COVID-19 positive subjects in the second step. Our experiments on the Track-2 dataset of the Second 2021 DiCOVA Challenge achieved the second top ranking with an AUC score of 81.21 and the top F1 score of 53.21 on a Blind Test set, improving the challenge baseline by 8.43% and 23.4% respectively and showing deployability, robustness and competitiveness with the state-of-the-art systems.

Keywords: Signal pattern classification, Data mining and big data methods - Pattern recognition, Physiological systems modeling - Signal processing in physiological systems
Abstract: In the context of monitoring patients with heart failure conditions, the automated assessment of heart sound quality is of major importance to insure the relevance of the medical analysis of the heart sound data. We propose in this study a technique of quality classification based on the selection of a small set of representative features. The first features are chosen to characterize whether the periodicity, complexity or statistical nature of the heart sound recordings. After segmentation process, the latter features are probing the detectability of the heart sounds in cardiac cycles. Our method is applied on a novel subcutaneous medical implant that combines ECG and accelerometric-based heart sound measurements. The actual prototype is in pre-clinical phase and has been implanted on 4 pigs, which anatomy and activity constitute a challenging environment for obtaining clean heart sounds. As reference quality labeling, we have performed a three-class manual annotation of each recording, qualified as ``good'', ``unsure'' and ``bad''. Our method allows to retrieve good quality heart sounds with a sensitivity and an accuracy of 82%+/-2% and 88%+/-6% respectively. Clinical relevance: By accurately recovering high quality heart sound sequences, our method will enable to monitor reliable physiological indicators of heart failure complications such as decompensation.

Keywords: Computer modeling for treatment planning, Neuromuscular systems - Stroke therapy devices / technologies
Abstract: Clinical outcome prediction plays an important role in stroke patient management where the modified Rankin scale (mRS), a measure of global disability, is widely applied.From a machine learning point-of-view, one of the main challenges is dealing with heterogeneous data at patient admission, i.e. the image data which are multi-dimensional and the clinical data which are scalars. In this paper, a multimodal convolutional neural network - long short-term memory (CNN-LSTM) based ensemble model is proposed.For each MR image, a dedicated network module provides preliminary prediction of the mRS score. The final mRS score is obtained by merging the preliminary probabilities of each module dedicated to a specific type of MR image weighted by the clinical metadata, here age or the NIH Stroke Scale/Score. The experimental results demonstrate that the proposed model surpasses the baselines and offers an original way to automatically encode the spatio-temporal context of MR images in a deep learning architecture. The highest AUC(0.77) was achieved for the proposed model with NIHSS.

Keywords: Diagnostic devices - Physiological monitoring, Ambulatory Diagnostic devices - Point of care technologies
Abstract: Autism spectrum disorder (ASD) is a neurodevelopmental condition that impacts language, communication and social interactions. The current diagnostic process for ASD is based upon a detailed multidisciplinary assessment. Currently no clinical biomarker exists to help in the diagnosis and monitoring of this condition that has a prevalence of approximately 1%. The electroretinogram (ERG), is a clinical test that records the electrical response of the retina to light. The ERG is a promising way to study different neurodevelopmental and neurodegenerative disorders, including ASD. In this study, we have proposed a machine learning based method to detect ASD from control subjects using the ERG waveform. We collected ERG signals from 47 control (CO) and 96 ASD individuals. We analyzed ERG signals both in the time and the spectral domain to gain insight into the statistically significant discriminating features between CO and ASD individuals. We evaluated the machine learning (ML) models using a subject independent cross validation-based approach. Time-domain features were able to detect ASD with a maximum 65% accuracy. The classification accuracy of our best ML model using time-domain and spectral features was 86%, with 98% sensitivity. Our preliminary results indicate that spectral analysis of ERG provides helpful information for the classification of ASD.

Keywords: Wireless technologies for interrogation of implantable therapeutic devices, Neuromuscular systems - Deep brain stimulation technologies
Abstract: Sensing technology, as well as cloud communication, is enabling the development of closed-loop deep brain stimulation for Parkinson’s disease. The accelerometer is a practical sensor that can provide information about the disease/health state of the patient as well as physical activity levels, all of which in the long-term can provide feedback information to an adaptive closed-loop control algorithm for more effective and personalized DBS therapy.

In this paper, we present for the first time, acceleration streamed from Medtronic’s RC+S device in patients with Parkinson’s disease while at home, and compare it to acceleration acquired concurrently from the patient’s Apple Watch. We examined the correlation between the accelerometer signals at varying time scales. We also compared the spectral band power obtained from the two accelerometers.

While there was an average correlation of 0.37 for subject 1 and 0.50 for subject 2 between the two acceleration signals on a time scale of 10 minutes, the correlation was lower for shorter time scales on the order of seconds. There was greater spectral power in the Parkinsonian tremor band of 4-7 Hz for the externally worn accelerometer than the internal accelerometer, but the internal accelerometer showed greater relative power distributed in the higher frequencies (7-30 Hz).

Thus, based on this preliminary analysis, we expect that the internal accelerometer may be used to assess patient activity and state for closed-loop DBS but tremor detection may require more sophisticated signal processing. Furthermore, the internal accelerometer may contain information in higher frequency bands that reveal information about the patient state.

Keywords: Neuromodulation devices
Abstract: In neuromodulation applications, conventional current mode stimulation is often preferred over its voltage mode equivalent due to its good control of the injected charge. However, it comes at the cost of less energy-efficient output stages. To increase energy efficiency, recent studies have explored non-rectangular stimuli. The current work highlights the importance of an adaptive supply for an output stage with programmable non-rectangular stimuli and accordingly proposes a system-level architecture for multi-channel stimulators. In the proposed architecture, a multi-output DC/DC Converter (DDC) allows each channel to choose among the available supply levels (i.e., DDC outputs) independently and based on its instant voltage/current requirement. A system-level analysis is carried out in Matlab to calculate the possible energy savings of this solution, compared to the conventional approach with a fixed supply. The energy savings have been simulated for a variety of supply levels and waveform amplitudes, suggesting energy savings of up to 83% when employing 6 DDC outputs and the lowest current amplitude explored (250 µA), and as high as 26% for a full-scale amplitude (4 mA).

Keywords: Neuromuscular systems - Deep brain stimulation technologies, Neuromuscular systems - Neural stimulation, Neuromuscular systems - Muscle stimulation
Abstract: Abstract — In this work, a stream function inverse boundary element method (IBEM) has been used for designing different deep transcranial magnetic stimulation (dTMS) coils to activate the prefrontal cortex and the temporal lobe have been set as the target regions. In addition, the performances of these coils have been described and the electric field induced by them has been obtained by using a computational forward technique. These results show that the stream function IBEM is an ideal approach to design optimal dTMS coils capable of producing deep stimulation in the target brain regions.

Clinical relevance — The design problem proposed here can be used to produce efficient dTMS stimulators for neurological disorders, which can overcome some of the currently existing limitations of the most common devices employed in TMS.

Keywords: General and theoretical informatics - Statistical data analysis, Health Informatics - Virtual reality in medicine
Abstract: Reliably measuring fear perception could help evaluate the effectiveness of treatments for pathological conditions such as specific phobias or post-traumatic stress syndrome (e.g., exposure therapy). In this study, we developed a novel virtual reality (VR) scenario to induce fear and evaluate the related physiological response by the analysis of skin conductance (SC) signal. Eighteen subjects voluntarily experienced the fear VR scenario while their SC was recorded. After the experiment, each participant was asked to score the perceived subjective fear using a Likert scale from 1 to 10. We used the cvxEDA algorithm to process the collected SC signals and extract several features able to estimate the autonomic response to the fearful stimuli. Finally, the extracted features were linearly combined to model the subjective fear perception scores by means of LASSO linear regression. The sparsification imposed by the LASSO procedure to mitigate the overfitting risk identified an optimal linear model including only the standard deviation of the tonic SC component as a regressor (p = 0.007; R² = 0.3337). The significant contribution of this feature to the model suggests that subjects experiencing more intense subjective fear have a more variable and unstable sympathetic tone.

Keywords: Health Informatics - Behavioral health informatics, Health Informatics - Healthcare communication networks, Health Informatics - Personal health systems
Abstract: A flexible meander-line monopole antenna (MMA) is presented for wireless medical body area networks. The antenna can be worn for on-and off-body healthcare applications. The overall dimension of the MMA is 37mm×50mm×2.37mm3. The MMA was manufactured and measured, and results matched with simulations. The MMA design shows a bandwidth of up to 1282.4 (450.5) MHz and provides gains of 3.03 (4.85) dBi in the lower and upper operating bands, respectively, showing omnidirectional radiation patterns in free space. While worn on the chest or arm, bandwidths as high as 688.9 (500.9) MHz and 1261.7 (524.2) MHz, and the gains of 3.80 (4.67) dBi and 3.00 (4.55) dBi were observed. The experimental measurements of the read range confirmed the results of the coverage range of up to 11 meters.

Keywords: Health Informatics - Clinical information systems, Health Informatics - Health data acquisition, transmission, management and visualization, Health Informatics - Health information systems
Abstract: Abstract—Subarachnoid hemorrhage (SAH) is a devastating neurological injury that can lead to many downstream complications including epilepsy. Predicting who will get epilepsy in order to find ways to prevent it as well as stratify patients for future interventions is a major challenge given the large number of variables not only related to the injury itself, but also to what happens after the injury. Extensive multimodal data are generated during the process of SAH patient care. In parallel, preclinical models are under development that attempt to imitate the variables observed in patients. Computational tools that consider all variables from both human data and animal models are lacking and demand an integrated, time-dependent platform where researchers can aggregate, store, visualize, analyze, and share the extensive integrated multimodal information. We developed a multi-tier web-based application that is secure, extensible, and adaptable to all available data modalities using flask micro-web framework, python, and PostgreSQL database. The system supports data visualization, data sharing and downloading for offline processing. The system is currently hosted inside the institutional private network and holds ~14 TB of data from 164 patients and 71 rodents.

Clinical Relevance—Our platform supports clinical and preclinical data management. It allows users to comprehensively visualize patient data and perform visual analytics. These utilities can improve research and clinical practice for subarachnoid hemorrhage and other brain injuries.

Keywords: Health Informatics - Telehealth, Sensor Informatics - Data inference, mining, and trend analysis, Health Informatics - Health data acquisition, transmission, management and visualization
Abstract: We present a cloud-based multimodal dialogue platform for the remote assessment and monitoring of speech, facial and fine motor function in Parkinson’s Disease (PD) at scale, along with a preliminary investigation of the efficacy of the various metrics automatically extracted by the platform. 22 healthy controls and 38 people with Parkinson's Disease (pPD) were instructed to complete four interactive sessions, spaced a week apart, on the platform. Each session involved a battery of tasks designed to elicit speech, facial movements and finger movements. We find that speech, facial kinematic and finger movement dexterity metrics show statistically significant differences between controls and pPD. We further investigate the sensitivity, specificity, reliability and generalisability of these metrics. Our results offer encouraging evidence for the utility of automatically-extracted audiovisual analytics in remote monitoring of PD and other movement disorders.

Keywords: Health Informatics - Telemedicine, Health Informatics - internet of things in healthcare, Health Informatics - eHealth
Abstract: In the last years there have been significant improvements in the accuracy of real-time 3D skeletal data estimation software. These applications based on convolutional neural networks (CNNs) can play a key role in a variety of clinical scenarios, from gait analysis to medical diagnosis. One of the main challenges is to apply such intelligent video analytic at a distance, which requires the system to satisfy, beside accuracy, also data privacy. To satisfy privacy by default and by design, the software has to run on ”edge” computing devices, by which the sensitive information (i.e., the video stream) is elaborated close to the camera while only the process results can be stored or sent over the communication network. In this paper we address such a challenge by evaluating the accuracy of the state-of-the-art software for human pose estimation when run ”at the edge”. We show how the most accurate platforms for pose estimation based on complex and deep neural networks can become inaccurate due to subsampling of the input video frames when run on the resource constrained edge devices. In contrast, we show that, starting from less accurate and ”lighter” CNNs and enhancing the pose estimation software with filters and interpolation primitives, the platform achieves better real- time performance and higher accuracy with a deviation below the error tolerance of a marker-based motion capture system.

Keywords: Sensor Informatics - Behavioral informatics, Sensor Informatics - Physiological monitoring, Sensor Informatics - Wearable systems and sensors
Abstract: Emotional computing has been previously applied to assess physiological behavior in a wide variety of tasks and activities. This study extends for the first time the use of emotional computing in the field of balance rehabilitation training. A proof-of-concept study was conducted to assess arousal and pleasure response to a range of physical exercises from the OTAGO and HOLOBALANCE balance rehabilitation programs with varying levels of difficulty and physical demand. Eleven participants were enrolled and performed a set of exercises wearing an ECG sensor, reporting arousal and pleasure at the end of each session. A dataset of 264 unique sessions was collected and used to extract heart rate variability (HRV) features from the measured RR intervals and automatically assess user arousal and pleasure, evaluating different classification algorithms. The results suggested that assessment of both emotions is feasible, reaching an accuracy of 72% and 74% for arousal and pleasure estimation, respectively. Clinical Relevance— Arousal and pleasure are clinically useful indicators of patient’s experience and engagement while performing balance rehabilitation exercises with novel sensing technologies and monitoring platforms.

Keywords: Robotic-aided therapies - Image guided surgery systems/ technologies, Clinical engineering -Verification and validation of diagnostic & therapeutic systems / technologies, Robotic-aided therapies - Computer-assisted surgery systems
Abstract: Optical tracking systems combined with imaging modalities such as computed tomography and magnetic resonance imaging are important parts of image guided surgery systems. By determining the location and orientation of surgical tools relative to a patient's reference system, tracking systems assist surgeons during the planning and execution of image guided procedures. Therefore, knowledge of the tracking system-induced error is of great importance. To this end, this study compared one passive and two active optical tracking systems in terms of their Target Registration Error. Two experiments were performed to measure the systems' accuracy, testing the impact of factors such as the size of the measuring volume, length of surgical instruments and environmental conditions with orthopedic procedures in mind. According to the performed experiments, the active systems achieved significantly higher accuracy than the tested passive system, reporting an overall accuracy of 0.063 mm (SD = 0.025) and 0.259 mm (SD = 0.152), respectively.

Keywords: Robotic-aided therapies - Computer-assisted surgery systems, Robotic-aided therapies - Surgical robotic systems, Robotic-aided therapies - Planning and execution technologies in surgical robotics
Abstract: Patient-specific templates (PST) have become a useful tool for guiding osteotomy in complex surgical scenarios such as pelvic resections. The design of the surgical template results in sharper, less jagged resection margins than freehand cuts. However, their correct placement can become difficult in some anatomical regions and cannot be verified during surgery. Conventionally, pelvic resections are performed using Computer Assisted Surgery (CAS), and in recent years Augmented Reality (AR) has been proposed in the literature as an additional tool to support PST placement. This work presents an AR task to simplify and improve the accuracy of the positioning of the template by displaying virtual content. The focus of the work is the creation of the virtual guides displayed during the AR task. The system was validated on a patient-specific phantom designed to provide a realistic setup. Encouraging results have been achieved. The use of the AR simplifies the surgical task and optimizes the correct positioning of the cutting template: an average error of 2.19 mm has been obtained, lower than obtained with state-of-the-art solutions. In addition, supporting PST placement through AR guidance is less time-consuming than the standard procedure that solely relies on anatomical landmarks as reference.

Keywords: Therapeutic devices and systems - ablation systems and technologies, Computer modeling for treatment planning
Abstract: We present a data-assimilation Bayesian framework in the context of laser ablation for the treatment of cancer. For solving the nonlinear estimation of the tissue temperature evolving during the therapy, the Unscented Kalman Filter (UKF) predicts the next thermal status and controls the ablation process, based on sparse temperature information. The purpose of this paper is to study the outcome of the prediction model based on UKF and to assess the influence of different model settings on the framework performances. In particular, we analyze the effects of the time resolution of the filter and the number and the location of the observations.

Keywords: Tactile displays and perception, Biologically inspired robotics and micro-biorobotics - Machine learning and control, Robot-aided surgery - Remote surgery systems / telesurgery
Abstract: Abstract—Researchers have adopted mechanistic and learning-based approaches for tip force estimation on soft robotic catheters. Typically, the literature attributes the mechanistic methods with more accuracy while indicating the learning-based methods outpace in computational time. In this study, a previously validated mechanistic tip force estimation method was compared with four learning-based methods, i.e., support vector- regression (SVR), random-forest (RF), AdaBoost (Ada), and deep neural network (DNN). The learning-based methods were trained on experimental data acquired from a robotic catheter, developed in-house. The accuracy of force estimation using the five methods were compared with the ground truth forces in a teleoperated catheter manipulation test. Moreover, the capability of the learning-based models in contact detection, i.e., detection of the onset of tip contact, were compared with the ground truth. The results showed that the mechanical model had a mean-absolute error (MAE) of 8.8 mN while the MAE of SVR, RF, Ada, and DNN were 5.6, 5.2, 5.3, and 5.1 mN, respectively. Moreover, the accuracy and precision of the mechanistic model for contact detection was 89.2% and 91.7%, respectively, while these were 97.0%, 97.7%, 97.6%, and 97% and 97.9%, 98.3%, 97.8%, and 98.8% for the SVR, RF, Ada, and DNN, respectively. The comparison showed that with hyperparameter optimization the learning-based models surpassed the mechanistic model in accuracy and precision, while both method approaches revealed acceptable performance for the proposed application.

Keywords: Machine learning / Deep learning approaches, Multimodal imaging, Magnetic resonance imaging - MR breast imaging
Abstract: Abstract— Segmentation of the thoracic region and breast tissues is crucial for analyzing and diagnosing the presence of breast masses. This paper introduces a medical image segmentation architecture that aggregates two neural networks based on the state-of-the-art nnU-Net. Additionally, this study proposes a polyvinyl alcohol cryogel (PVA-C) breast phantom, based on its automated segmentation approach, to enable planning and navigation experiments for robotic breast surgery. The dataset consists of multimodality breast MRI of T2W and STIR images obtained from 10 patients. A statistical analysis of segmentation tasks emphasizes the Dice Similarity Coefficient (DSC), segmentation accuracy, sensitivity, and specificity. We first use a single class labeling to segment the breast region and then exploit it as an input for three-class labeling to segment fatty, fibroglandular (FGT), and tumorous tissues. The first network has a 0.95 DCS, while the second network has a 0.95, 0.83, and 0.41 for fat, FGT, and tumor classes, respectively.

Keywords: Image segmentation, Machine learning / Deep learning approaches
Abstract: Semantic segmentation of anatomical structures in laparoscopic videos is a crucial task to enable the development of new computer-assisted systems that can assist surgeons during surgery. However, this is a difficult task due to artifacts and similar visual characteristics of anatomical structures on the laparoscopic videos. Recently, deep learning algorithms have been showed promising results on the segmentation of laparoscopic instruments. However, due to the lack of large public datasets for semantic segmentation of anatomical structures, there are only a few studies on this task. In this work, we evaluate the performance of five networks, namely U-Net, U-Net++, DynUNet, UNETR and DeepLabV3+, for segmentation of laparoscopic cholecystectomy images from the recently released CholecSeg8k dataset. To the best of our knowledge, this is the first benchmark performed on this dataset. Training was performed with dice loss. The networks were evaluated on segmentation of 8 anatomical structures and instruments, performance was quantified through the dice coefficient, intersection over union, recall, and precision. Apart from the U-Net, all networks obtained scores similar to each other, with the U-Net++ being the network with the best overall score with a mean Dice value of 0.62. Overall, the results show that there is still room for improvement in the segmentation of anatomical structures from laparoscopic videos.

Keywords: Magnetic resonance imaging - MR neuroimaging, Brain imaging and image analysis, Machine learning / Deep learning approaches
Abstract: The APOE-ε4 allele is a known genetic risk for Alzheimer’s disease (AD). Thus, it can be reasoned that the APOE-ε4 allele would also impact neurodegeneration-associated structural brain changes. Here we probe if the APOE-ε4 genotype directly modulates the human brain’s gray matter using a neural network trained on the whole-brain gray matter images from the cognitively normally aging (CN) and AD individuals. To investigate the linkage between the APOE-ε4 allele and whole-brain (voxel-wise) gray matter, we systematically profile our investigation in multiple classification tasks, including diagnostic classification and APOE-ε4 classification conjointly as well as independently. Results suggest that although the MRI data can reliably track and reflect neurodegenerative changes in the brain cross-sectionally, the APOE-ε4 status may not be distinguishable correspondingly. The nonexistence of a direct and convincing modulative effect of APOE-ε4 on the whole-brain gray matter indicates that the gray matter changes may be independent of the APOE-ε4 status, and instead characterize a non-APOE, comorbid mechanism in AD.

Keywords: Deformable registration, Multimodal image fusion, Machine learning / Deep learning approaches
Abstract: Many applications in image-guided surgery and therapy require fast and reliable non-linear, multi-modal image registration. Recently proposed unsupervised deep learning-based registration methods have demonstrated superior performance compared to iterative methods in just a fraction of the time. Most of the learning-based methods have focused on mono-modal image registration. The extension to multi-modal registration depends on the use of an appropriate similarity function, such as the mutual information (MI). We propose guiding the training of a deep learning-based registration method with MI estimation between an image-pair in an end-to-end trainable network. Our results show that a small, 2-layer network produces competitive results in both mono- and multi-modal registration, with sub-second run-times. Comparisons to both iterative and deep learning-based methods show that our MI-based method produces topologically and qualitatively superior results with an extremely low rate of non-diffeomorphic transformations. Real-time clinical application will benefit from a better visual matching of anatomical structures and less registration failures/outliers.

Keywords: Functional image analysis, Optical imaging - Confocal microscopy, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Interstitial Cells of Cajal (ICC) are specialized gastrointestinal (GI) pacemaker cells that generate and actively propagate slow waves of depolarization (SWs) of the muscularis propria. SWs regulate the motility of the GI tract necessary for digestion, absorption of nutrients, and elimination of waste. Within the gastric wall, there are three main inter-connected layers of ICC networks: longitudinal muscle ICC (ICC-LM), myenteric plexus ICC (ICC-MP) & circumferential muscle (ICC-CM). Fractal structural parameters such as Fractal Dimension (FD), Lacunarity and Succolarity, have many advantages over physically-based parameters when it comes to characterizing the complex architectures of ICC networks. The analysis of networks of ICC throughout the proximal and distal murine gastric antrum with the FD and Lacunarity metrics was previously performed. Although the application of Succolarity is relatively nascent compared to the FD and Lacunarity; nevertheless, numerous studies have demonstrated the capability of this fractal measure to extract information from images associated with flow by which neither the FD nor Lacunarity are capable of discerning. In this study, Succolarity analysis of ICC-MP and ICC-CM networks were performed with confocal images taken across the proximal and distal murine antrum. Our findings demonstrated the Succolarity of ICC-MP and ICC-CM varied with directions and antral regions. The Succolarity of ICC-MP did not vary considerably with direction, however, Succolarity was higher in the aboral direction with 0.2113  0.1589, and 0.0637  0.0822 in the proximal and distal antrum, respectively. The overall Succolarity of ICC-MP was significantly higher than that of ICC-CM in the proximal antrum (0.1580 ± 0.1325 vs 0.0008 ± 0.0007, p-value=0.0008) and in the distal antrum (0.0449 ± 0.0409 vs 0.0006 ± 0.0010, p-value=0.003).

Keywords: Infra-red imaging, Image feature extraction, Image analysis and classification - Machine learning / Deep learning approaches
Abstract: Onchocerciasis is causing blindness in over half a million people in the world today. Drug development for the disease is crippled as there is no way of measuring effectiveness of the drug without an invasive procedure. Drug efficacy measurement through assessment of viability of onchocerca worms requires the patients to undergo nodulectomy which is invasive, expensive, time-consuming, skill-dependent, infrastructure dependent and lengthy process. In this paper, we discuss the first-ever study that proposes use of machine learning over thermal imaging to non-invasively and accurately predict the viability of worms. The key contributions of the paper are (i) a unique thermal imaging protocol along with pre-processing steps such as alignment, registration and segmentation to extract interpretable features (ii) extraction of relevant semantic features (iii) development of accurate classifiers for detecting the existence of viable worms in a nodule. When tested on a prospective test data of 30 participants with 48 palpable nodules, we achieved an Area Under the Curve (AUC) of 0.85.

Clinical Relevance— This is the first ever research effort of using thermal imaging in the assessment of viability of onchocerca worms and it resulted in a very high specificity>95% which makes it a promising modality to pursue further.

Keywords: Multiscale image analysis, Image analysis and classification - Digital Pathology, Machine learning / Deep learning approaches
Abstract: We present a multi-scale graphical network that can capture the relevant representations of individual cell morphology, topological structure of cell communities in a tissue image, as well as whole slide level attributes. This helps to effectively merge the disease relevant cell morphology to the overall topological context within the sample, within one unified deep framework. From the explainability point of view, instead of empirical design, the graphs are designed with biomedical considerations in mind in order to have translational validity. We also provide a clinically interpretable visualisation of the cells and their micro- and macro-environment by leveraging label noise reduction. We demonstrate the efficacy of our methodology on myeloproliferative neoplasms (MPN), a haematopoietic stem cell disorder as an exemplar test case. The proposed method achieves an encouraging performance in the robust separation of different MPN subtypes in this exciting new dataset as part of this work.

Keywords: X-ray radiography, Image segmentation, Machine learning / Deep learning approaches
Abstract: Automatic lesion segmentation in mammography images assists in the diagnosis of breast cancer, which is the most common type of cancer especially among women. The robust segmentation of mammography images has been considered a backbreaking task due to: i) the low contrast of the lesion boundaries; ii) the extremely variable lesions’ sizes and shapes; and iii) some extremely small lesions on the mammogram image. To overcome these drawbacks, Deep Learning methods have been implemented and have shown impressive results when applied to medical image segmentation. This work presents a benchmark for breast lesion segmentation in mammography images, where six state-of-the-art methods were evaluated on 1692 mammograms from a public dataset (CBIS-DDSM), and compared considering the following six metrics: i) Dice coefficient; ii) Jaccard index; iii) accuracy; iv) recall; v) specificity; and vi) precision. The base U-Net, UNETR, DynUNet, SegResNetVAE, RF-Net, MDA-Net architectures were trained with a combination of the cross-entropy and Dice loss functions. Although the networks presented Dice scores superior to 86%, two of them managed to distinguish themselves. In general, the results demonstrate the efficiency of the MDA-Net and DynUnet with Dice scores of 90.25% and 89.67%, and accuracy of 93.48% and 93.03%, respectively.

Keywords: Neural signals - Machine learning & Classification, Brain functional imaging - Mapping, Neural interfaces - Microelectrode technology
Abstract: Identifying different functional regions during a brain surgery is a challenging task usually performed by highly specialized neurophysiologists. Progress in this field may be used to improve in situ brain navigation and will serve as an important building block to minimize the number of animals in preclinical brain research required by properly positioning implants intraoperatively. The study at hand aims to correlate recorded high-density extracellular signals with the volume of origin by deep learning methods. Our work establishes connections between the position in the brain and recorded high-density neural signals. This was achieved by evaluating the performance of BLSTM, BGRU, QRNN and CNN neural network architectures on multisite electrophysiological data sets. All networks were able to successfully distinguish cortical and thalamic brain regions according to their respective neural signals. The GRU provides the best results with an accuracy of 88.6 % and demonstrates that this classification task might be solved in higher detail while minimizing complex preprocessing steps.

Keywords: Motor learning, neural control, and neuromuscular systems, Brain physiology and modeling - Neural dynamics and computation
Abstract: Implanted microelectrode arrays can directly pick up electrode signals from the primary motor cortex (M1) during movement, and brain-machine interfaces (BMIs) can decode these signals to predict the directions of contemporaneous movements. However, it is not well known how much each individual input is responsible for the overall performance of a BMI decoder. In this paper, we seek to quantify how much each channel contributes to an artificial neural network (ANN)-based decoder, by measuring how much the removal of each individual channel degrades the accuracy of the output. If information on movement direction was equally distributed among channels, then the removal of one would have a minimal effect on decoder accuracy. On the other hand, if that information was distributed sparsely, then the removal of specific information-rich channels would significantly lower decoder accuracy. We found that for most channels, their removal did not significantly affect decoder performance. However, for a subset of channels (16 out of 61), removing them significantly reduced the decoder accuracy. This suggests that information is not uniformly distributed among the recording channels. We propose examining these channels further to optimize BMIs more effectively, as well as understand how M1 functions at the neuronal level.

Keywords: Neurological disorders, Neurological disorders - Diagnostic and evaluation techniques, Neurorehabilitation
Abstract: Parkinson is the second most common neurodegenerative disease, mainly related to progressive locomotor alterations caused by dopamine deficiency. The gait kinematic is a principal disease biomarker that associates patterns like the step length, flexed posture, and bradykinesia with disease progression. Nonetheless, these patterns are analyzed from invasive setups that only capture coarse dynamics at advanced disease stages. This work introduces a very compact and robust deep representation that effectively learns a Riemannian manifold to describe locomotor parkinsonian patterns. Contrary to traditional deep strategies, the presented framework fully explores data geometry from input mean covariance matrices representing video sequences. These symmetric positive definite (SPD) matrices lie in a Riemannian manifold. Then such matrices are projected to the SPD net, which learns a bank of SPD matrices from a non-linear training, that may be exploited in a hierarchical composition from a set of layers. In a final layer, a projection in a Euclidean space allows learning the discriminatory patterns of PD w.r.t control population. In a retrospective study with a total of 22 patients (11 Parkinson's and 11 controls), the proposed approach achieves a remarkable classification between Parkinson's and control video sequences, correctly labeling all Parkinson's patients, and outperforming typical 3D convolutional representations.

Keywords: Brain physiology and modeling, Neural signals - Machine learning & Classification, Brain functional imaging - EEG
Abstract: The complexity of brain activity involved in the generation of the experience of pain makes it hard to identify neural markers able to predict pain states. The within and between subjects variability of pain hinders the predictive potential of machine learning models trained across participants. This challenge can be tackled by implementing deep learning classifiers based on convolutional neural networks (CNNs). We targeted phase-based connectivity in the alpha band recorded with electroencephalography (EEG) during resting states and sensory conditions (eyes open [O] and closed [C] as resting states, and warm [W] and hot [H] water as sensory conditions). Connectivity features were extracted and re-organized as square matrices, because CNNs are effective in detecting the patterns from 2D data. To assess the classifier performance we implemented two complementary approaches: we 1) trained and tested the classifier with data from all participants, and 2) using a leave-one-out approach, that is excluding one participant at a time during training while using their data as a test set. The accuracy of binary classification between pain condition (H) and eyes open resting state (O) was 94.16% with the first approach, and 61.01% with the leave-one-out approach.

Keywords: Neural signals - Machine learning & Classification, Neurological disorders - Traumatic brain injury, Neural signal processing
Abstract: Machine learning and deep learning algorithms have paved the way for improved analysis of biomedical data which has led to a better understanding of various biological conditions. However, one major hindrance to leveraging the potential of machine learning models is the requirement of huge datasets. In the biomedical domain, this becomes extremely difficult due to uncertainties in collecting high-quality data as well as, in the case of human subjects' data, privacy. Further, when it comes to biomedical data, inter-subject variability has been a long-entrenched issue. The data obtained from different individuals will differ to a considerable extent that it becomes difficult to find population differences in small datasets. In this work, we investigate the use of label alignment techniques on an EEG-based Traumatic Brain Injury (TBI) classification task to overcome inter-subject variability, thereby increasing the classification accuracy. We show an increase in accuracy of around 6% in some cases as compared to our previous results. In the end, we also propose a methodology to incorporate TBI data from a different species (e.g., mice) after domain adaptation, which might further improve the performance by increasing the amount of training datasets available for the classification model.

Keywords: Brain-computer/machine interface, Neural signals - Machine learning & Classification
Abstract: Ideal brain-computer interfaces (BCIs) need to be efficient and accurate, demanding for classifiers that can work across subjects while providing high classification accuracy results from recordings with short duration. To address this problem, we present a new deep learning framework for discriminating motor imagery (MI) tasks from electroencephalography (EEG) signals. The framework consists of a 1D convolutional neural network-long short-term memory (CNN-LSTM), combined with a dynamic channel selection approach based on Davies-Bouldin index (DBI). Using data from BCI competition IV-IIa data, the proposed framework reports an average classification accuracy of 70.17% and 76.18% when using only 800 ms and 1500 ms of the EEG data after the task onset, respectively. The proposed framework dynamically balances individual differences, achieves comparable or better performance compared to existing work, while using short duration of EEG.