CHASE '23: Proceedings of the 8th ACM/IEEE International Conference on Connected Health: Applications, Systems and Engineering Technologies

Full Citation in the ACM Digital Library

Parkinson's Disease Action Tremor Detection with Supervised-Leaning Models

People with Parkinson's Disease (PD) have multiple symptoms, such as freezing of gait (FoG), hand tremors, speech difficulties, and balance issues, in different stages of the disease. Among these symptoms, hand tremors are present across all stages of the disease. PD hand tremors have critical consequences and negatively impact the quality of PD patients' everyday lives. Researchers have proposed a variety of wearable devices to mitigate PD tremors. However, these devices require accurate tremor detection technology to work effectively while the tremor occurs. This paper introduces a PD action tremor detection method to recognize PD tremors from regular activities. We used a dataset from 30 PD patients wearing accelerometers and gyroscope sensors on their wrists. We selected time-domain and frequency-domain hand-crafted features. Also, we compared our hand-crafted features with existing CNN data-driven features, and our features have more specific boundaries in 2-D feature visualization using the t-SNE tool. We fed our features into multiple supervised machine learning models, including Logistic Regression (LR), K-Nearest Neighbours (KNNs), Support Vector Machines (SVMs), and Convolutional Neural Networks (CNNs), for detecting PD action tremors. These models were evaluated with 30 PD patients' data. The performance of all models using our features has more than 90% of F1 scores in five-fold cross-validations and 88% F1 scores in the leave-one-out evaluation. Specifically, Support Vector Machines (SVMs) perform the best in five-fold cross-validation with over 92% F1 scores. SVMs also show the best performance in the leave-one-out evaluation with over 90% F1 scores.

Social Visual Behavior Analytics for Autism Therapy of Children Based on Automated Mutual Gaze Detection

Social visual behavior, as a type of non-verbal communication, plays a central role in studying social cognitive processes in interactive and complex settings of autism therapy interventions. However, for social visual behavior analytics in children with autism spectrum disorder (ASD), it is challenging to manually collect and evaluate gaze data due to the human coder's time and effort costs. In this paper, we introduce a social visual behavior analytics approach by quantifying the mutual gaze performance of children receiving play-based autism interventions using an automated mutual gaze detection framework. Our analysis is based on a video dataset that captures and records social interactions between children with autism and their therapy trainers (28 observations, 84 video clips, 21 hrs duration). The effectiveness of our framework was evaluated by comparing the mutual gaze ratio derived from the mutual gaze detection framework with the human-coded ratio values. We analyzed the mutual gaze frequency and duration across different therapy settings, activities, and sessions. We created mutual gaze-related measures for social visual behavior score prediction using multiple machine learning-based regression models. The results show that our method provides mutual gaze measures that reliably represent the human coders' hand-coded social gaze measures and effectively evaluates and predicts ASD children's social visual performance during the intervention. Our findings have implications for social interaction analysis in small-group behavior assessments in numerous co-located settings in (special) education and in the workplace.

FoG-Finder: Real-time Freezing of Gait Detection and Treatment

Freezing of gait is a serious symptom of Parkinson's disease that increases the risk of injury through falling, and reduces quality of life. Current clinical freezing of gait treatments fail to adequately address the fall risk posed by freezing of gait symptoms, and current real-time treatment systems have high false positive rates. To address this problem, we designed a closed-loop, non-intrusive, and real-time freezing of gait detection and treatment system, FoG-Finder, that automatically detects and treats freezing of gait. To evaluate FoG-Finder, we first collected 716 freezing of gait events from 11 patients. We then compared FoG-Finder against other realtime systems with our dataset. Our system was able to achieve a 13.4% higher F1 score and a 10.7% higher overall accuracy while achieving a reduction of 85.8% in the false positive treatment rate compared with other validated real-time freezing of gait detection and treatment systems. Additionally, FoG-Finder achieved an average treatment latency of 427ms and 615ms for subject-dependent and leave-one-subject-out settings, respectively, making it a viable system to treat freezing of gait in the real-world.

EMS-BERT: A Pre-Trained Language Representation Model for the Emergency Medical Services (EMS) Domain

Emergency Medical Services (EMS) is an important domain of healthcare. First responders save millions of lives per year. Machine learning and sensing technologies are actively being developed to support first responders in their EMS activities. However, there are significant challenges to overcome in developing these new solutions. One of the main challenges is the limitations of existing methods for EMS text mining, and developing a highly accurate language model for the EMS domain. Several important Bidirectional Encoder Representations from Transformer (BERT) models for medical domains, i.e., BioBERT and ClinicalBERT, have significantly influenced biomedical text mining tasks. But extracting information from the EMS domain is a separate challenge due to the uniqueness of the EMS domain, and the significant scarcity of a high-quality EMS corpus. In this research, we propose EMS-BERT - a BERT model specifically developed for EMS text-mining tasks. For data augmentation on our small, classified EMS corpus which consists of nearly 2.4M words, we use a simultaneous pre-training method for transfer-learning relevant information from medical, bio-medical, and clinical domains; and train a high-performance BERT model. Our thorough evaluation shows at least 2% to as much as 11% improvement of F-1 scores for EMS-BERT on different classification tasks, i.e., entity recognition, relation extraction, and inferring missing information when compared both with existing state-of-the-art clinical entity recognition tools, and with various medical BERT models.

Active Reinforcement Learning for Personalized Stress Monitoring in Everyday Settings

Most existing sensor-based monitoring frameworks presume that a large available labeled dataset is processed to train accurate detection models. However, in settings where personalization is necessary at deployment time to fine-tune the model, a person-specific dataset needs to be collected online by interacting with the users. Optimizing the collection of labels in such phase is instrumental to impose a tolerable burden on the users while maximizing personal improvement. In this paper, we consider a fine-grain stress detection problem based on wearable sensors targeting everyday settings, and propose a novel context-aware active learning strategy capable of jointly maximizing the meaningfulness of the signal samples we request the user to label and the response rate. We develop a multilayered sensor-edge-cloud platform to periodically capture physiological signals and process them in real-time, as well as to collect labels and retrain the detection model. We collect a large dataset and show that the context-aware active learning technique we propose achieves a desirable detection performance using 88% and 32% fewer queries from users compared to a randomized strategy and a traditional active learning strategy, respectively.

DOVE: Shoulder-based Opioid Overdose Detection and Reversal Device

Naloxone is a life-saving drug capable of reversing a fatal opioid overdose. Although this drug has existed for over 50 years, opioid overdose-related deaths have consistently risen and surpassed 120,000 globally in 2021. Opioids induce respiratory depression by activating μ-opioid receptors at specific sites in the central nervous system. This results in overdose deaths caused by slow and shallow breathing, also known as opioid-induced respiratory depression. 1.6 million individuals suffer from opioid use disorder annually, making them at high risk of overdose, primarily due to the increasing prevalence of Fentanyl. Over 52% of these deaths occur when the individual is alone. Immediate response to an overdose by delivering naloxone can save the individual's life. To solve this problem, we developed a closed-loop sensor-driven auto-injector that can determine a fatal overdose and inject naloxone. 76% of this population is willing to wear such a device on the shoulder, a canonical injection site. This paper presents the DOVE, a shoulder-based opioid overdose detection and reversal device. It noninvasively measures the subject's motion state and changes in blood oxygen levels (SpO2) along with the respiration state. These biomarkers are measured from the shoulder using an optical sensor and accelerometer to determine if a fatal overdose occurred. We evaluated our DOVE device against an FDA-cleared commercial pulse oximeter by inducing apneic events as they have very similar SpO2 trends to an overdose. Results show that SpO2 can be measured on the shoulder across different skin tones with an accuracy of 96.8% and a high Pearson correlation of 0.766 (p < 0.0001).

Detecting Eating, and Social Presence with All Day Wearable RGB-T

Social presence has been known to impact eating behavior among people with obesity; however, the dual study of eating behavior and social presence in real-world settings is challenging due to the inability to reliably confirm the co-occurrence of these important factors. High-resolution video cameras can detect timing while providing visual confirmation of behavior; however, their potential to capture all-day behavior is limited by short battery lifetime and lack of autonomy in detection. Low-resolution infrared (IR) sensors have shown promise in automating human behavior detection; however, it is unknown if IR sensors contribute to behavior detection when combined with RGB cameras. To address these challenges, we designed and deployed a low-power, and low-resolution RGB video camera, in conjunction with a low-resolution IR sensor, to test a learned model's ability to detect eating and social presence. We evaluated our system in the wild with 10 participants with obesity; our models displayed slight improvement when detecting eating (5%) and significant improvement when detecting social presence (44%) compared with using a video-only approach. We analyzed device failure scenarios and their implications for future wearable camera design and machine learning pipelines. Lastly, we provide guidance for future studies using low-cost RGB and IR sensors to validate human behavior with context.

Using Geographic Location-Based Public Health Features in Survival Analysis

Time elapsed till an event of interest is often modeled using the survival analysis methodology, which estimates a survival score based on the input features. There is a resurgence of interest in developing more accurate prediction models for time-to-event prediction in personalized healthcare using modern tools such as neural networks. Higher quality features and more frequent observations improve the predictions for a patient, however, the impact of including a patient's geographic location-based public health statistics on individual predictions has not been studied. This paper proposes a complementary improvement to survival analysis models by incorporating public health statistics in the input features. We show that including geographic location-based public health information results in a statistically significant improvement in the concordance index evaluated on the Surveillance, Epidemiology, and End Results (SEER) dataset containing nationwide cancer incidence data. The improvement holds for both the standard Cox proportional hazards model and the state-of-the-art Deep Survival Machines model. Our results indicate the utility of geographic location-based public health features in survival analysis.

Virtual Therapy Exergame for Upper Extremity Rehabilitation Using Smart Wearable Sensors

Virtual reality (VR) has been utilized for several applications and has shown great potential for rehabilitation, especially for home therapy. However, these systems solely rely on information from VR hand controllers, which do not fully capture the individual movement of the joints. In this paper, we propose a creative VR therapy exergame for upper extremity rehabilitation using multidimensional reaching tasks while simultaneously capturing hand movement from the VR controllers and elbow joint movement from a non-invasive fabric-based wearable sensor made by coating knit fabric with carbon nanotube. We conducted a preliminary study with non-clinical participants (n = 12, 7 F). In a 2 × 2 within-subjects study (orientation (vertical, horizontal) × configuration (flat, curved)), we evaluated the effectiveness and enjoyment of the exergame. The results show that there was a statistically significant difference in terms of task completion time between the two orientations. However, no significant differences were found in the number of mistakes in both orientation and configuration of the virtual exergame. This can lead to customizing therapy while maintaining the same level of intensity. The results of the resistance change generated from the wearable sensor revealed that the flat configuration in the vertical orientation induced more elbow stretches than the other conditions. Finally, we reported the subjective measures based on questionnaires for usability and user experience in different study conditions. In conclusion, the proposed VR exergame has the potential as a multi-modal sensory tool for personalized upper extremity home-based therapy and telerehabilitation.

Efficient and Direct Inference of Heart Rate Variability using Both Signal Processing and Machine Learning

Heart Rate Variability (HRV) measures the variation of the time between consecutive heartbeats and is a major indicator of physical and mental health. Recent research has demonstrated that photoplethysmography (PPG) sensors can be used to infer HRV. However, many prior studies had high errors because they only employed signal processing or machine learning (ML), or because they indirectly inferred HRV, or because there lacks large training datasets. Many prior studies may also require large ML models. The low accuracy and large model sizes limit their applications to small embedded devices and potential future use in healthcare.

To address the above issues, we first collected a large dataset of PPG signals and HRV ground truth. With this dataset, we developed HRV models that combine signal processing and ML to directly infer HRV. Evaluation results show that our method had errors between 3.5% to 25.7% and outperformed signal-processing-only and ML-only methods. We also explored different ML models, which showed that Decision Trees and Multi-level Perceptrons have 13.0% and 9.1% errors on average with models at most hundreds of KB and inference time less than 1ms. Hence, they are more suitable for small embedded devices and potentially enable the future use of PPG-based HRV monitoring in healthcare.

Interpreting High Order Epistasis Using Sparse Transformers

Genome-Wide Association Studies aim to identify relations between Single Nucleotide Polymorphisms (SNPs) and the manifestation of certain diseases, which is an important challenge in biomedicine and personalized healthcare. However, most genetic diseases are explained by the interactions between several SNPs, known as epistasis. Detecting epistasis is a very computationally demanding task, due to the sheer number of SNP combinations to analyze. Recently, deep learning has emerged as a possible solution for genomic prediction, but the black-box nature of neural networks and the lack of explainability is a drawback yet to be solved. In this paper, a new, flexible framework for interpreting neural networks for anyorder epistasis detection is presented. Using sparse transformers, a technique not yet employed for epistasis detection, different SNP representations are explored and attention scores are assigned to each SNP to quantify its relevance for phenotype prediction. The results on simulated datasets show that the proposed framework outperforms state-of-the-art methods for explainability, identifying SNP interactions in diverse epistasis scenarios. The proposed framework is validated on a real breast cancer dataset, identifying second to fifth order interactions in the top 40% most relevant SNPs.

Exploring Earables to Monitor Temporal Lack of Focus during Online Meetings to Identify Onset of Neurological Disorders

This paper presents a framework called enGauge that leverages ear-based inertial sensing to continuously monitor listener focus levels in online meetings and provide feedback to the speaker about audience engagement. This allows for the identification of the onset of several neurodevelopmental disorders, including attention deficit hyperactivity disorder (ADHD), and can help to improve the effectiveness of online meetings by allowing speakers to adjust their speaking pace and style based on audience engagement. We explore a contrastive learning-based approach coupled with a judicial selection of anchor events from the meeting contents to model the system. enGauge can detect patterns or shifts in behavior and focus levels of passive listeners to accurately identify changes in focus. Results from a user study with 38 participants showed an overall F1-score of 0.89 for detecting passive listeners' focus levels. Our study suggests that ear-based inertial sensing has the potential to be a valuable tool for the early detection and monitoring of several neurodevelopmental disorders among individuals.

Short: Real-Time Bladder Monitoring by Bio-impedance Analysis to Aid Urinary Incontinence

The inability to sense the need to urinate is a persistent trouble for many individuals who have urinary incontinence. There are no products for people with urinary incontinence and their caregivers that target the elimination of involuntary urination instead of containing leaks or warning about leaks. Thus, fulfilling the real-time monitoring of the bladder to notify urination time holds a life-changing innovation potential for people with urinary incontinence. In this paper, we propose a low-cost wearable device in the form factor of an unobtrusive belt that uses bio-impedance measurement across the bladder region to notify the patients or caregivers about the need to urinate. We also present results from human subject tests using the proposed wearable prototype with a custom machine-learning algorithm to evaluate the accuracy of the system. Results from the human-subject tests showed an accuracy of over 90% on the binary task of full versus empty bladder states based on the change in bio-impedance values.

Short: Racial Disparities in Pulse Oximetry Cannot Be Fixed With Race-Based Correction

Pulse oximeters play a critical role in health monitoring. Pulse ox measurements have statistical bias that is a function of race, which results in higher rates of occult hypoxemia, i.e., missed detection of dangerously low oxygenation, in patients of color. This paper further characterizes the statistical distribution of pulse ox measurements, showing they also have a higher variance for patients racialized as Black, compared to those racialized as white. By analyzing the performance of hypoxemia detection as a detector, we show that no single race-based correction factor will provide equal performance by race. As a result, for racially equitable pulse oximetry, the pulse oximeter itself must be fixed, not just the hypoxemia thresholds.

Short: Integrated Sensing Platform for Detecting Social Isolation and Loneliness In the Elderly Community

Social isolation is an objective absence or paucity of contacts and interactions between a person and a social network. Loneliness is a subjective feeling of being alone, separated, or apart from others. These two highly correlated mental health concerns significantly increase the elderly's risk of premature death from all causes, a risk that may rival those of smoking, obesity, and physical inactivity. In this work, we propose an integrated design for technology usage to tackle the aforementioned mental healthcare concerns involving the elderly community. The technology infrastructure is an Internet-of-Things (IoT) monitoring system aiming to detect social isolation and promptly predict the risk of loneliness. With our system, interventions can be developed to maintain the mental well-being of the elderly, a salient need due to the social disruptions incurred by COVID-19.

Short: Basal-Adjust: Trend Prediction Alerts and Adjusted Basal Rates for Hyperglycemia Prevention

Significant advancements in type 1 diabetes treatment have been made in the development of state-of-the-art Artificial Pancreas Systems (APS). However, lapses currently exist in the timely treatment of unsafe blood glucose (BG) levels, especially in the case of rebound hyperglycemia. We propose a machine learning (ML) method for predictive BG scenario categorization that outputs messages alerting the patient to upcoming BG trends to allow for earlier, educated treatment. In addition to standard notifications of predicted hypoglycemia and hyperglycemia, we introduce BG scenario-specific alert messages and the preliminary steps toward precise basal suggestions for the prevention of rebound hyperglycemia. Experimental evaluation on the DCLP3 clinical dataset achieves >98% accuracy and >79% precision for predicting rebound high events for patient alerts.

Short: RF-Q: Unsupervised Signal Quality Assessment for Robust RF-based Respiration Monitoring

Continuous monitoring of respiration provides invaluable insights about health status management (e.g., the progression or recovery of diseases). Recent advancements in radio frequency (RF) technologies show promise for continuous respiration monitoring by virtue of their non-invasive nature, and preferred over wearable solutions that require frequent charging and continuous wearing. However, RF signals are susceptible to large body movements, which are inevitable in real life, challenging the robustness of respiration monitoring. While many existing methods have been proposed to achieve robust RF-based respiration monitoring, their reliance on supervised data limits their potential for broad applicability. In this context, we propose, RF-Q, an unsupervised/self-supervised model to achieve signal quality assessment and quality-aware estimation for robust RF-based respiration monitoring. RF-Q uses the reconstruction error of an autoencoder (AE) neural network to quantify the quality of respiratory information in RF signals without the need for data labeling. With the combination of the quantified signal quality and reconstructed signal in a weighted fusion, we are able to achieve improved robustness of RF respiration monitoring. We demonstrate that, instead of applying sophisticated models devised with respective expertise using a considerable amount of labeled data, by just quantifying the signal quality in an unsupervised manner we can significantly boost the average end-to-end (e2e) respiratory rate estimation accuracy of a baseline by an improvement ratio of 2.75, higher than the gain of 1.94 achieved by a supervised baseline method that excludes distorted data.

Short: Precision Polysubstance Use Episode Detection in Wearable Biosensor Data Streams

Wearable biosensors create the opportunity for continuous health monitoring by generating streams of measurements that reflect users' physiological conditions in natural environments. Continuous health monitoring is a key enabler of precision health, a means to detect individual-level changes early and initiate personalized preventive measures or other clinical interventions. Although the amount of data generated is theoretically unbounded, precise labeling is rare outside of controlled clinical environments. Using data streams from a study of patients recovering from cocaine use disorder, we demonstrate early results of a novel method to detect polysubstance use without precisely labeled training data using an anomaly detection paradigm. RP-STREAM performed better than an alternative in detecting polysubstance use in wearable biosensor data streams. The proposed semi-supervised learning model makes efficient use of training data and computational resources while also automating parameter selection. We also identify the effects of THC and cocaine polysubstance use in wearable biosensor data.

Short: Deep Learning Approach to Skeletal Performance Evaluation of Physical Therapy Exercises

At-home exercising strongly predicts physical therapy patient outcomes, underscoring the need for analyzing patient behaviors at-home via remote patient monitoring. Contemporary methods for remote patient monitoring rely on specialized sensors, i.e., Inertial Measurement Units, RGB-Depth cameras, motion capture systems, or stereo vision which are costly and not scalable to all physical therapy patients. Here, we observe a lack of literature using only a monocular RGB camera. In this paper, we demonstrate a skeletal feedback model for at-home exercises using only video acquired from a smartphone camera. We propose models for (i) Patient Performance Evaluation - which classifies the correctness of exercises, and (ii) Guidance - which identifies why the exercise went wrong so the patient can correct themselves. We use these models on our dataset of four common physical therapy exercises labeled by a physical therapist. Our results demonstrate the feasibility of using skeletal data from state-of-the-art 3D human pose estimation models for physical rehabilitation exercise evaluation and guidance. Thus, we enable remote patient monitoring and guidance from a single camera - making it highly cost-effective and scalable.

Demo: P-Fall: Personalization Pipeline for Fall Detection

We present an edge-cloud collaborative personalized fall detection pipeline called P-Fall. A personalized fall detection model requires real-time adaptation of a pre-trained model using real-time feedback data provided by the user. We, herein, highlight the design of the software architecture for a collaborative framework, the smart-watch's UI for the ease of collecting a user's feedback data, and the automation of the personalization process.

Demo: Addressing Inter-Intra Patient Variability via Personalized Meta-Federated Learning in IoT-Enabled Health Monitoring

Federated learning (FL) has been widely adopted in IoT-enabled health monitoring on biosignals. However, the global model may not adapt well to each target patient's data due to complex biosignals' morphological characteristics caused by inter- and intra-patient variability. To address the challenge, we propose a personalized meta-federated learning framework (PMFed) for patient-specific health monitoring in IoT. Experimentally, we evaluate the effectiveness and generalization of PMFed over three health monitoring tasks on a physical IoT platform. Experimental results show that the PMFed excels at empirical performances when compared with SOTA personalized FL algorithms.

Tala Box: an Interactive Embedded System to Accompany Patients with Cognitive Disorders

Given the lack of drugs to treat cognitive deficit disorders, non-pharmacological approaches have become popular to address these pathologies. In particular, music therapy has been successfully used with Alzheimer's patients to improve their mood and functional abilities. The Tala Sound project studies whether non-Western music from the southern regions of India can be used to improve the mood and functional abilities of Western people suffering from Alzheimer's disease. This traditional style of music, called Carnatic music, is based on specific rhythm sequences, called "talas", that have an irregular rhythm compared to the ones found in Western music.

In this paper, we describe the Tala Box, a new multimedia sensory device with which patients can interact while a Carnatic song is being played. The motivation for the development of this embedded system is the intent, in addition to the unusual nature of the music played, to use motor and cognitive stimulation to better engage Alzheimer's patients and, hopefully, improve their mood and mental state. Preliminary testing with healthy seniors has shown great interest in the Tala Box.

Poster: Foot-Floor Friction Based Walking Surface Detection for Fall Prevention Using Wearable Motion Sensors

Automatic walking surface detection helps people adapt their gait to different surfaces and reduce fall risk. Walking on different surfaces causes different foot-floor friction patterns. We proposed to deploy motion sensors near the ankle to sense foot-floor friction and recognize walking surfaces. There are two contributions in this proposed research work. First, we demonstrated that the proposed method is capable of distinguishing five most-common walking surfaces in daily living. Second, we compare the detection accuracy between walking normally and dragging feet while walking. Experimental results show the proposed method obtains higher accuracy for dragging feet while walking, which reaches 90.6% using only five seconds of data.

Poster: Machine Learning based Real Time Detection of Freezing of Gait of Parkinson Patients Running on a Body Worn Device

For those who have Parkinson's disease, one of the most incapacitating symptoms is Freezing of Gait (FOG). Gait impairment and disruptions limit everyday activities and reduce quality of daily life along with the increase in the risk of falling [1]. Thanks to recent advancement in embedded electronics and sensors as well as their adaptation in the wearable device market, low power devices are becoming more and more capable running neural networks. This enables researchers to implement complex models on wearable devices that capture and analyze sensor data to detect FOG in real-time.

Poster: Virtual Reality Exergame for Upper Extremity Rehabilitation Using Smart Wearable Sensors

In this work, we propose a creative VR therapy exergame with multi-dimensional reaching tasks for upper extremity rehabilitation. Our system tracks data from the upper extremities using VR hand controllers and a flexible carbon nanotube sensor positioned on the elbow. We conducted a preliminary study (n=12, 7 F) to evaluate the exergame's therapeutic factors, including orientation (horizontal, vertical), configuration (flat, curved), and user experience. The results show a statistically significant difference in task completion time for the variable orientation, but no significance for the number of mistakes. For the resistance change generated from the carbon nanotube sleeve, the flat configuration in the vertical orientation significantly induced more elbow stretches than the other conditions. These results suggest that VR therapy can be customized to the patient without a change in the intensity and induced body stretch. Our proposed VR exergame has the potential to personalize upper extremity home-based therapy using multi-modal sensory data collection.

Poster: Design of a Music Intervention System Using Social Robotics for Cognitive Enhancement

To aid in the care of older adults living with Alzheimer's disease and related dementias (ADRD), many forms of therapies and interventions have been created. One such approach, music intervention, uses song and dance to engage users cognitively, physically, and emotionally. This form of intervention has been proven effective for persons living with ADRD, yet it has not been widely implemented due to a low number of professionals (caregivers, music intervention specialists, etc.) performing the interventions. Social robots can provide relief to caregivers by engaging older adults living with ADRD in music. Therefore, we suggest using social robots to use song and dance to interact with older adults living with ADRD through the Music intervention Using Socially Engaging robotics (MUSE) app. This app will have older adults with ADRD participate in a variety of music-based activities that promote positive cognitive, physical, and emotional well-being.

Poster: Quantifying Signal Quality Using Autoencoder for Robust RF-based Respiration Monitoring

While radio frequency (RF) based respiration monitoring for at-home health screening is receiving increasing attention, robustness remains an open challenge. In recent work, deep learning (DL) methods have been demonstrated effective in dealing with nonlinear issues from multi-path interference to motion disturbance, thus improving the accuracy of RF-based respiration monitoring. However, such DL methods usually require large amounts of training data with intensive manual labeling efforts, and frequently not openly available. We propose RF-Q for robust RF-based respiration monitoring, using self-supervised learning with an autoencoder (AE) neural network to quantify the quality of respiratory signal based on the residual between the original and reconstructed signals. We demonstrate that, by simply quantifying the signal quality with AE for weighted estimation we can boost the end-to-end (e2e) respiration monitoring accuracy by an improvement ratio of 2.75 compared to a baseline.

Poster: A Distributed Deep Reinforcement Learning System for Medical Image Segmentation

Multi-institutional collaboration is an emerging deployment of medical imaging processing with the goal to address the scarce annotation problem. While most of the efforts in this domain focus on the supervised machine learning models and the model performance improvement, there lack the discussion about the distributed system performance, such as the trade-off between collaboration and efficiency, i.e., communication cost and processing time. In this work, we propose a distributed system based on deep reinforcement learning for medical image segmentation. Preliminary experiments are conducted on single and multiple CPU and GPU environments to demonstrate the system performance and the trade-off. We highlight some insights for better designs of multi-institutional collaboration in the future.

Poster: Towards Robust, Extensible, and Scalable Home Sensing Data Collection

Home-based health monitoring systems are important to many conditions (e.g., aging, chronic diseases). The absence of suitable data collection infrastructure is a fundamental barrier to the development of related algorithms and systems. In this poster, we present Proteus, a robust, extensible and scalable data collection infrastructure, to enable small research teams to manage large deployments. We identify the desired features and achieve them by combining mature technologies and new components:i) extensibility with new, diverse sensor types and data formats with a few lines of coding (LOC) efforts; ii) scalability in managing sensor/edge devices to automate many deployment, management tasks; iii) resilience to system failures and network outage. Experiments on a prototype show zero data loss or system error for one sensor node running 10 days, and 99.95% of data received for 32 emulated sensors sending data at 200 Mbps, 20 and 100 fold reductions in node setup efforts and LOC for new sensor types. The preliminary results show Proteus is promising for large-scale longitudinal deployment of home-based health monitoring.

Poster: Highly Nonlinear Solitary Wave Transducers for Detecting Eye Pressure Changes

Utilizing a highly nonlinear solitary wave (HNSW) transducer, it is possible to sense the intraocular pressure (IOP) of an eye. A miniature transducer that captures HNSWs has the potential to be utilized at the point-of-care for patients and for self administered monitoring of IOP. It is possible to mimic a human eye using a Polydimethylsiloxane (PDMS) cornea that is pressurized to 10mmhg, 20mmhg, and 30mmhg. The HNSW response, when tested on PDMS at the three different pressures, are unique waves with distinct features. These features can be extracted from each signal and compared to verify whether HNSWs provide enough differentiation between IOPs which would allow for correct identification.

Poster: BioFactCheck: Exploring the Feasibility of Explainable Automated Inconsistency Detection in Biomedical and Health Literature

There are inconsistencies in conclusions drawn from the studies that address the same research question in the biomedical literature. This paper presents preliminary work on the approaches taken to build an inconsistency detection and explanation model starting with the development of a gold-standard contradiction sentences corpus. First, we utilize SemRep, a third-party tool that can automatically segment any biomedical sentence into the form of a subject, predicate, and object. A pair of sentences with the same subject/object but different predicates is identified as contradictory sentences. These sentences are then manually curated by domain experts to filter out noise. In the future, we plan to generate a large manually curated gold-standard contradiction sentence dataset and use that for developing an automated tool for detecting and extracting contradictions in biomedical and health text.

Development of a Custom Wrist Wearable for Use in Nursing Homes

The wearable technology market is consistently growing. Many devices are being developed for commercial applications and individual use incorporating sensors that measure vital information such as heart rate, blood oxygen, and temperature. These general market application devices lack optimization for some target markets such as nursing home patients. For this application, optimized hardware and software are required to extract meaningful data to improve outcomes on a wide scale. Patient-maintained devices are not a practical option; thus, the task of charging and setup is placed on Certified Nursing Assistants (CNAs). These wearables are wireless sensors without user displays and controls. Typical sensing devices consist of Inertial Measurement Units (IMUs), temperature sensors, and photoplethysmography (PPG) sensors for heart rate (HR) and SpO2 with wireless communication and a rechargeable battery. The key objectives for specific use in nursing homes are optimal hardware design for measurement and power efficiency and firmware for the measurement scenarios present in the nursing home. An overview of HR, SpO2, IMU, and temperature measurements will be presented with unique considerations for the target population. The battery life of this wrist wearable was estimated to last 79 days.

Poster: Noninvasive Respirator Fit Factor Inference by Semi-Supervised Learning

The need for personal protective equipment, such as respirators, has been emphasized by pandemics as they provide protection against infectious diseases. Adequate protection is only possible when respirators fit properly and are worn correctly. Therefore, it is especially critical to closely monitor and ensure respirator fit, particularly during a pandemic. To ensure proper fit and continuous monitoring, we propose a new noninvasive method that uses speech signals to measure the attenuation of sound caused by the respirator. This method provides a quantitative measure of respirator Fit Factor (FF, the ratio of the concentration of a substance in ambient air to its concentration inside the respirator). This method is also cost-effective and easy to implement. By collecting limited labeled and unlabeled speech data, augmenting labeled data, extracting time and frequency domain features, we achieved up to 86.24% accuracy in respirator fit detection using semi-supervised learning model.

Poster: Automatic Compliance Analysis on Clinical Notes and Lifestyle Guidelines in Cancer Survivorship

Maintaining a healthy lifestyle has been proven to have significant benefits in cancer survivorship. It is expected that clinicians have a comprehensive understanding of publicly published cancer lifestyle guidelines and can effectively convey the information to patients. The objective of this paper is to develop an automatic text analysis method to assess the compliance of lifestyle information provided during medical visits and publicly available guidelines. Preliminary results show that selected lifestyle keywords appear an average of 3.54 times per medical visit, and 7% of medical notes pertain to patients' lifestyle. Semantic analysis and word dictionary will be applied to evaluate the extent of information compliance and inform strategies for improving lifestyle recommendations for cancer survivors.

Poster: Design of Mixed Reality Dangerous Situations for Autistic Children: Road Safety

While many road safety programs have been instituted over the past decades, bringing down the rate of injury for children from 3.6 deaths per 100,000 to .3, still, every year in the United States, approximately 600 child pedestrians are killed and 76,000 child pedestrians are injured. Children on the autistic spectrum often require additional considerations than neuro-typical children when trying to teach road safety skills, including differences in thinking flexibility, difficulty in understanding social contexts and communication, and over-sensitivity or under-sensitivity to sights and sounds. A mixed reality road safety environment design is proposed to address several considerations, with potential technical and clinical design challenges.

AsyncFedKD: Asynchronous Federated Learning with Knowledge Distillation

Federated learning (FL) allows for the decentralized training of a global model on edge devices without transferring data samples, thus preserving privacy. Due to the ubiquitous wearable devices and mobile devices with health applications, FL has shown promise in the medical field for applications such as medical imaging, disease diagnosis, and electronic health record (EHR) analysis. However, slower edge devices with limited resources can slow down the training process. To address this issue and increase efficiency, we propose the use of Asynchronous Federated Learning with Knowledge Distillation (AsyncFedKD). AsyncFedKD asynchronously trains a lightweight global student model using a pre-trained teacher model, preventing a decrease in training efficiency due to slow edge devices. The knowledge distillation aspect of AsyncFedKD effectively compresses the size of model parameters for efficient communication during training. AsyncFedKD has been tested on a sensitive mammography cancer dataset and achieved an accuracy of 88% on the global model.

Probing the EHR for Standardized Nursing Data

Nursing documentation is essential for the welfare of patients and for productive communication between healthcare professionals. Currently, nursing care is documented by means of standardized and specific non-standardized nursing terminologies that various healthcare companies provide. Because of significant differences between terminologies, nursing professionals devote considerable time to map distinct terminologies by manually searching terminology databases or books. We present an automated approach that finds mappings between terminologies of two widely-used nursing care plans: it is based on UMLS as an intermediate resource, and on similarity computed via language models. According to our nursing team experts, our best-performing model found accurate mappings for approximately 54 percent of terms.

Poster: Affordable Automatic Air Quality Monitoring System for At-risk Homes

This research aims to improve indoor environmental justice (EJ) outcomes for elderly individuals residing in historically under-resourced communities of color in Knoxville, TN. To achieve this goal, we have created an autonomous air quality monitoring system that continuously tracks the temperature and CO2 concentrations inside the building. The user interface will display alert messages to remind users to take the essential precautions to ensure their health and safety in the event of hazardous conditions. In addition, the system includes a notification mechanism that alerts loved ones and neighbors to step in when a risk cannot be properly mitigated by a single person. In Knoxville's historically under-resourced communities of color, we successfully implemented and tested our autonomous air quality monitoring system.

Enabling Data Interoperability for Decentralized, Smart, and Connected Health Applications

The digitization of healthcare is increasing, providing a wealth of data that can improve patient care and outcomes. However, different systems and formats often silo this data, presenting a significant challenge to achieving a connected and interoperable healthcare ecosystem. Data interoperability is crucial for realizing the full potential of smart and connected health, which can improve patient outcomes and reduce costs. In this paper, we describe the key concepts and technologies enabling interoperability, including health information exchange, interoperability standards, and application programming interfaces enabling connected health applications built on blockchains. We also discuss the challenges and barriers to achieving data interoperability, such as privacy and security concerns, organizational and cultural barriers, and technical limitations. Furthermore, we highlight the benefits of data interoperability in healthcare, such as better patient outcomes, improved population health management, reduced costs, and increased efficiency. Finally, we discuss the potential future of data interoperability in healthcare, including the role of emerging privacy-preserving technologies.