The application of machine and deep learning algorithms in Human Activity Recognition (HAR) has shown great potential for monitoring various professional and daily life activities, benefiting different research areas such as healthcare, well-being and industrial automation. HAR can enable the development of various services and applications to empower technical performance and enable risk prevention in working places, to support education and training, and, more in general, to monitor the biopsychosocial status of people. However, we still lack a baseline framework for easily implementing the data processing pipeline that must be designed to setup and configure HAR workflows. This makes challenging to estimate the effectiveness, efficiency, and the overall quality of HAR solutions, thus hindering the comparison among different approaches. This also increases the likelihood that researchers introduce errors, which negatively affect the accuracy of the obtained results. To fill in the gap, this paper introduces B-HAR, an open-source framework to automatically implement baseline HAR workflows.
Both diet and physical activity are associated with obesity and chronic diseases such as diabetes and metabolic syndrome. Early efforts in connecting dietary and physical activity behaviors to generate patterns rarely considered the use of time. In this paper, we propose a distance-based cluster analysis approach to find joint temporal diet and physical activity patterns among U.S. adults ages 20-65. Dynamic Time Warping (DTW) generalized to multi-dimensions is combined with commonly used clustering methods to generate unbiased partitioning of the National Health and Nutrition Examination Survey 2003-2006 (NHANES) dataset. The clustering results are evaluated using visualization of the clusters, the Silhouette Index, and the associations between clusters and health status indicators based on multivariate regression models. Our experiments indicate that the integration of diet, physical activity, and time has the potential to discover joint temporal patterns with association to health.
Automatic sleep staging based on electroencephalography (EEG) and electromyography (EMG) signals is an important aspect of sleep-related research. Current sleep staging methods suffer from two major drawbacks. First, there are limited information interactions between modalities in the existing methods. Second, current methods do not develop unified models that can handle different sources of input. To address these issues, we propose a novel sleep stage scoring model sDREAMER, which emphasizes cross-modality interaction and per-channel performance. Specifically, we develop a mixture-of-modality-expert (MoME) model with three pathways for EEG, EMG, and mixed signals with partially shared weights. We further propose a self-distillation training scheme for further information interaction across modalities. Our model is trained with multi-channel inputs and can make classifications on either single-channel or multi-channel inputs. Experiments demonstrate that our model outperforms the existing transformer-based sleep scoring methods for multi-channel inference. For single-channel inference, our model also outperforms the transformer-based models trained with single-channel signals.
The emergence of COVID-19 offered a unique opportunity to study chronic pain patients as they responded to sudden changes in social environments, increased community stress, and reduced access to care. We report findings from n=70 Spinal Cord Stimulation (SCS) patients before and during initial pandemic stages resulting from advances in home monitoring and artificial intelligence that produced novel insights despite pandemic-related disruptions. From a multi-dimensional array of frequently monitored signals—including mobility, sleep, voice, and psychological assessments—we found that while the overall patient cohort appeared unaffected by the pandemic onset, patients had significantly different individual experiences. Three distinct patient responses (sub-cohorts) were revealed, those with: worsened pain, reduced activities, or improved quality-of-life. Remarkably, none of the specific measures by themselves were significantly affected; instead, it was their synergy that exposed the effects elicited by the pandemic onset. Partial correlations illustrating linked dimensions by sub-cohort during the pandemic and those associations were different for each sub-cohort before COVID-19, suggesting that daily at-home telemonitoring of chronic conditions may reveal novel patient types. This work highlights the opportunities afforded by applying modern analytic techniques to more holistic and longitudinal patient outcomes, which might aid clinicians in making more informed treatment decisions in the future.
Digitized histopathology glass slides, known as Whole Slide Images (WSIs), are often several gigapixels large and contain sensitive metadata information, which makes distributed processing unfeasible. Moreover, artifacts in WSIs may result in unreliable predictions when directly applied by Deep Learning (DL) algorithms. Therefore, preprocessing WSIs is beneficial, e.g., eliminating privacy-sensitive information, splitting a gigapixel medical image into tiles, and removing the diagnostically irrelevant areas. This work proposes a cloud service to parallelize the preprocessing pipeline for large medical images. The data and model parallelization will not only boost the end-to-end processing efficiency for histological tasks but also secure the reconstruction of WSI by randomly distributing tiles across processing nodes. Furthermore, the initial steps of the pipeline will be integrated into the Jupyter-based Virtual Research Environment (VRE) to enable image owners to configure and automate the execution process based on resource allocation.
The classification of heart arrhythmias through recordings of its electrical activity (ECGs) represent a coveted non-invasive diagnostic tool for the detection of life threatening conditions. Nevertheless, the design of fast and effective automatic deep learning procedures solving this task is not trivial. In this work, we developed an automatic pipeline for heart rhythm classification relying on a lighter temporal representation based on a scalar vectorcardiogram (SVCG). The signal is preliminarly downsampled using a Fast Fourier Transform (FFT) retaining high-level features. Afterwards, each ECG is converted to a 3D compact representation through Inverse Dower’s transform and fed through a specffically designed convolutional neural network. The reliability of the proposed method is validated and compared to standard 12-lead ECGs from PhysioNet Computing in Cardiology Challenge (2020). We obtained a competitive test categorical accuracy for the largest dataset (SVCG 90.6-ECG S7.1 on PTBXL) and comparable results for the remaining sources. Compared with the existing methods, we propose an efficient method for the classification of heart arrhythmias through vectorcardiography, avoiding the need of hand-crafted features also having a lower computational and timing effort.
Healthcare-related research studies often deploy ecological momentary assessment techniques to sample information from participants in their natural environment. This paper presents an online system to support remote access of ecological self-report and actigraphy-based measurements for individuals with Amyotrophic Lateral Sclerosis (ALS) and their caregivers. The presented framework includes a custom mobile app and makes use of a web-based application programming interface for data collection with wrist-worn actigraphy devices. The system was evaluated with a research protocol to measure sleep, fatigue, and pain for individuals with ALS and their caregivers (N=8) over a consecutive seven-day period. Though daily self-report response rates were widespread (0%-100%) and the remote actigraphy collection varied in reliability, novel relationships between individuals with ALS and their caregivers were identified from the collected data. Online, ecological systems can support real-time remote monitoring and/or interventions to help understand diseases like ALS and advance healthcare research.
Descriptive pattern mining is a useful tool in expansion of knowledge. One such area of descriptive pattern mining is that of sequential pattern mining. In sequential mining, items maintain an order of occurrence. In this paper, we present a digital health solution for mining sequential patterns from real-life healthcare data. Specifically, it is a non-trivial extension to the sequential mining algorithm PrefixSpan. Through an association of time, we find improved relevance of a pattern overall significance relative to a focal point. This is particularly useful in the medical domain, where significance of information varies depending on the time of its occurrence. For example, consider a time of being diagnosed with a disease. A condition occurring 16 years prior to the time of diagnosis provides less information than the same condition occurring 2 years prior to diagnosis. In traditional sequential mining, both conditions would equally contribute to support, despite their unequal value in describing causes of diagnosis. To resolve such issue, we provide an inclusion of two additional user-defined parameters to incorporate time within itemsets-namely, a timeline interval (describing the length of an interval, of which itemsets of different intervals are treated separately by their difference in time to a focal point), as well as a maximal window (denoting the maximal interval that disallows for any greater time difference than such interval). With timelines associated to itemsets, relevance of itemsets have improved interpretability for domain experts.
An innovative building design technology is being modelled to kill all pathogens including COVID-19 inside the building naturally before it attacks the human body. In this study, a solar irradiance has been applied through an exterior glazing wall of the building to create ultraviolet germicidal irradiation (UVGI), which is derived from sunlight, to destroy all pathogens inside the buildings before they attack human bodies. Research shows that all pathogens, including COVID-19, can be destroyed by UVGI’s short-range wavelengths of 254–280 nanometers by rupturing their nucleic acid bonds, forcing them to malfunction their biochemical operations and ultimately causing the pathogen to die which indeed be an innovative field of science to eliminate pathogen naturally before it penetrates to the human body.
The anterior cruciate ligament (ACL) stabilizes the knee joint to prevent internal rotation. ACL injuries are common for athletes in high-cutting sports, affecting female athletes at a greater incidence compared to male athletes. Wearable devices and digital health technologies, broadly speaking, have become increasingly utilized in clinical trials to quantify patient reported outcome measures to complement subjective assessments. In the context of sports medicine, wearable technology serves as an objective and continuous means to complement athlete ratings of perceived exertion. One such biomarker of interest is muscle oxygen saturation (SmO 2 ), which is the ratio of oxygenated hemoglobin to total hemoglobin. A decrease in SmO 2 is indicative of greater muscle exertion, higher energy output, and greater oxygen consumption. Current assessments for ACL rehabilitation employ subjective means and lack the integration of continuous, internal data. This study bridges this gap via the measurement of SmO 2 to guide the return to play (RTP) process following ACL reconstruction (ACLR). Preliminary results from one patient demonstrate significant differences between the surgical and contralateral limbs during max-minute fan bike and Tabata fan bike exercises at both 6-and 9-month trials following ACLR. In the bilateral leg press exercise, significant differences were found between the surgical and contralateral lower extremities at 6-months but not at the 9-month trial. Data collection will also occur at 12-months post ACLR to further momtor differences between surgical and contralateral limbs in the RTP process. These results provide the impetus to enable the interoperability of data gathered from wearable devices into data management systems for optimizing performance and health of athletes following injury.
Photoplethysmography (PPG) is a non-invasive technique employed for the detection of blood volume fluctuations within microvessels, accomplished by measuring variations in light absorption. PPG provides valuable insights into several cardiovascular parameters, such as vascular age, stiffness index, and reflection index. By comparing these parameters with reference values from the literature, physicians can obtain a comprehensive assessment of the subject’s cardiovascular health.This paper presents the CardiaPPG instrument, specifically designed for the early screening and identification of cardio-vascular diseases through PPG analysis. The major features of CardiaPPG include its non-invasiveness, lightweight and portable nature, suitability for large-scale population screening, including low-income, rural, and developing regions, and user-friendliness even for non-specialized personnel. Additionally, the instrument establishes a correlation ($R^{2}$ = 0.94) between the calculated cardiovascular age using PPG and the individual’s demographic age. This correlation can potentially aid in the identification of early-onset severe cardiovascular conditions, such as hypertension, hypercholesterolemia, and ischemic heart disease.
Modelling biological pathway plays an important role in understanding different processes for decision making, especially in forensic investigations on doping activities in sports. Recently, the issue of sample swapping has arisen as a potential fraudulent behaviour by athletes to avoid a positive doping test result. The current detection models neglect an important factor, i.e., leveraging the steroid metabolism pathway of the human body. The spatial relationships between different metabolites within the steroid metabolism pathways are important and cannot be merely treated as linear correlations when assessing similarities among the samples obtained from athletes. To address this challenge, we propose the GRAMP model based on graph representation learning to incorporate domain knowledge into the model decision for the detection of sample swapping. Our model takes into account the spatial structural dependencies of different metabolites using a graph attention mechanism and generates high-level embeddings to detect fraudulent behaviour. We evaluate our approach through extensive experiments on real-world datasets and find that our proposed model outperforms existing state-of-the-art models for fraud detection tasks in sports, demonstrating the effectiveness of our approach and its potential impact on decision making.
So far, initial treatment recommendations for internet-based cognitive behavioral therapy (iCBT) decision support were mostly high-level or static. Personalized treatment recommendations could pave the way toward better treatment outcomes and adaptive treatments by leveraging information from past patients. We explore the disadvantages of multi-class recommendation and propose a modular approach using multilabel classification for treatment recommendations. Our machine learning-based treatment recommender composes treatment programs from a set of modules. It achieves a 79.02% F1-score on historically successful treatments, significantly outperforming the existing system by around 4% while offering other advantages such as interpretability and robustness. Using our recommendation as an initial starting point, clinicians can adjust the modular treatments to provide a more personalized treatment.
In healthcare industry, it is a standard practice to assign a set of International Classification of Diseases (ICD) to a clinical note (which can be a patient visit, a discharge summary and the like) as part of medical coding process mandated by medical care and patient billing. A supervised framework is adopted for most of the automated ICD coding assignment methods in which a subset of the clinical notes are a-priori labeled with ICD codes. But in lot of cases enough labeled texts are not available. These call for an unsupervised assignment of ICD codes. However, the quality of the data plays an important role in the performance of unsupervised coding, - low quality data leads to degradation of performance. In this paper, we explore a transfer learning approach for ICD coding using a combination of pre-training and supervised fine-tuning. We use a hierarchical BERT model comprising of a Bi-LSTM layered on top of BERT (this removes the restriction on the size of clinical texts)) as part of model architecture, and pre-train it on the total corpus (which include both labeled and unlabeled data). Next we transfer its weights to fine tune the model with labeled data (MIMIC data) in a supervised framework and then use this model to predict ICD code for unlabeled data using token similarity. This is the first use of using transfer learning in ICD prediction to our knowledge. Finally we show the efficacy of our transfer learning approach through rigorous experimentation, - there is 20% gain of sensitivity (recall) and 6% lift in specificity in ICD prediction compared to direct unsupervised prediction.
Healthcare delivery transformations and the use of connected health devices are paving the way to a paradigm shift from current healthcare systems towards patient-centered systems. Many proposals successfully reorient health information systems so that data are still distributed among the institutions and services that generate them, while being accessed jointly from a single point of view per patient. However, this means that control over the operations involving this data is lost. Mechanisms to maintain the traceability of health data are needed. This will enable the verification of the integrity of the records and will provide assurances that they have not been compromised. This problem has already been addressed in other domains such as food supply chains, where traceability allows to know all interactions with a food supply from the time it is produced until it is consumed. This paper proposes a blockchain solution to achieve the traceability of health data in patient-centered distributed environments. To validate this proposal, a case study involving 50 sociosanitary institutions in Portugal have been chosen. Different performance tests have been conducted to demonstrate the suitability, feasibility and scalability of our proposal.
An accurate prediction of blood glucose levels for individuals affected with type-1 diabetes mellitus helps to regulate blood glucose through specific insulin delivery. In our work, we propose the design of a densely-connected encoder-decoder network in conjunction with Long-Short Term Memory networks. We formulate the blood glucose prediction as a deep reinforcement learning problem and evaluate our results on the OhioT1DM dataset. The OhioT1DM dataset contains blood glucose monitoring records in intervals of 5 minutes over 8 weeks for 12 patients affected with type-1 diabetes mellitus. Prior works aim to predict the blood glucose levels in prediction horizons of 30 and 45 minutes, corresponding to 6 and 9 data points, respectively. Compared to prior work with the best prediction accuracy so far with respect to the mean absolute error, we improve by 18.4% and 22.5% in 30-minute and 45-minute prediction horizons, respectively. Furthermore, for risk assessment in our predictions, we visualize the error and evaluate clinical risk through a surveillance error grid approach.
The inefficiency of the healthcare system in addressing pandemics is highlighted after COVID-19 which is mostly rooted in data availability and accuracy. As it is believed we might witness more pandemics in future, our research's main objective is to propose an integrated health system to support healthcare preparedness for future infectious outbreaks and pandemics. The system could support managers and authorities in healthcare and disaster management, and policymakers through data collection, sharing, and analysis.
A significant percentage of university students reported poor sleep quality due to various factors. This affects their academic performance, mental well-being, and overall health. To address this, we propose a real-time monitoring system that can effectively identify students at risk of poor sleep quality. Our system utilizes two modes of continuous data collection (i.e., consumer-grade wearables and sleep diary) to inform both students and university well-being centers about sleep quality characteristics over the past week. Incorporating both subjective and objective measures facilitates a more holistic appraisal of sleep quality. We conducted a validation study in which we collected sleep health data from university students for four weeks to show how the proposed system could be implemented. The results of our study showed that it is feasible to implement the proposed system, although no agreement between the two modes of data collection was found. This calls for future research efforts aimed at devising a composite metric that harmoniously combines these two modes.
The fem-tech sector is still nascent and estimated to reach ${\$}$103 billion by 2030. It is likely to incorporate all themes related to women’s health and technology. In this sector, startups are growing amidst competition to stay relevant; nevertheless, their mortality rate is very high. This study identifies the factors that drive front-end innovation in fem-tech startups for their successful product development. We conducted a national-level survey to collect the views of various fem-tech stakeholders for empirical investigation, and the data were validated using Partial Least Square Structural Equation Modeling (PLS-SEM.) The Standardized Root Mean Square Residual (SRMR) was below 0.07, and the Normed Fit Index (NFI) was closer to 0.9, indicating a good model fit. The factors’ path model implies a successful fem-tech product development direction. It contributes to the literature on successful health venture creation and offers recommendations to sectoral stakeholders.
This study discusses the design and evaluation of a virtual reality (VR) game for the psychosocial support of adolescents with orofacial cleft lip and/or palate (CLP). In collaboration with global cleft organization, Smile Train, and psychologists from their international network of partner hospitals and clinics, we designed a 2-player VR game allowing psychologists and patients to work together on psychosocial skills development, focusing on the unique challenges adolescent patients with CLP face. The game was evaluated by a psychologist and patients over the course of a month, providing insight to whether a VR game would be feasible in providing psychosocial support to this population.
