Wireless Mobile Communication and Healthcare

Mobile devices contain an increasing number of sensors, many of which can be used for disease diagnosis and monitoring. Thus along with the ease of access and use of mobile devices there is a trend towards developing neurological tests onto mobile devices. Speech-based approaches have shown particular promise in detection of neurological conditions. However, designing such tools carries a number of challenges, such as how to manage noise, delivering the instructions for the speech based tasks, handling user error, and how to adapt the design to be accessible to specific populations with Parkinson’s Disease and Amyotrophic Lateral Sclerosis. This report discusses our experiences in the design of a mobile-based application that assesses and monitors disease progression using speech changes as a biomarker.

A schizophrenia relapse has severe consequences for a patient’s health, work, and sometimes even life safety. If an oncoming relapse can be predicted on time, for example by detecting early behavioral changes in patients, then interventions could be provided to prevent the relapse. In this work, we investigated a machine learning based schizophrenia relapse prediction model using mobile sensing data to characterize behavioral features. A patient-independent model providing sequential predictions, closely representing the clinical deployment scenario for relapse prediction, was evaluated. The model uses the mobile sensing data from the recent four weeks to predict an oncoming relapse in the next week. We used the behavioral rhythm features extracted from daily templates of mobile sensing data, self-reported symptoms collected via EMA (Ecological Momentary Assessment), and demographics to compare different classifiers for the relapse prediction. Naive Bayes based model gave the best results with an F2 score of 0.083 when evaluated in a dataset consisting of 63 schizophrenia patients, each monitored for up to a year. The obtained F2 score, though low, is better than the baseline performance of random classification (F2 score of 0.02 ± 0.024). Thus, mobile sensing has predictive value for detecting an oncoming relapse and needs further investigation to improve the current performance. Towards that end, further feature engineering and model personalization based on the behavioral idiosyncrasies of a patient could be helpful.

Depression is widespread and, despite a wide range of treatment options, causes considerable suffering and disease burden. Digital health interventions, including self-monitoring and self-management, are becoming increasingly important to offer e-mental health treatment and to support the recovery of people affected. SELFPASS is such an application designed for the individual therapy of patients suffering from depression. To gain more insights, this study aims to examine e-mental health treatment using the example of SELFPASS with two groups: healthy people and patients suffering from depression. The analysis includes the measurement of the constructs Usability, Trust, Task-Technology Fit, Attitude and Intention-to-use, the causal relationships between them and the differences between healthy and depressive participants as well as differences between participants’ evaluations at the beginning and at the end of the usage period. The results show that the Usability has the biggest influence on the Attitude and the Intention-to-use. Moreover, the study reveals clear differences between healthy and depressive participants and indicates the need for more efforts to improve compliance.

Human memory comprises one of the most complex brain functions, attracting researchers to unveil the neural mechanisms governing its effective operation. In this respect, the current study examines the application of a wearable single-channel EEG to the interpretation of cognitive operations reflecting memory processes. For this purpose, we implemented a set of tasks for evaluating the participants’ processing skills and memory efficiency, in order to examine potential outcomes derived from a specialized cognitive training routine. The employed training method targeted the distinction of automatic and controlled processing and its effects on memory, while we also investigated transfer effects to untrained tasks. Based on the electrophysiological data recorded during the cognitive tasks, we computed measures of induced EEG activity for each frequency band to examine the influence of cognitive training on both task performance and brain activity, as well as whether the EEG metrics could provide insight into the underlying brain processes and augment the interpretation of behavioral outcomes. Ultimately, statistical analysis showed an apparent contribution of EEG in understanding the observed behavioral differences, while our training program had a clear impact on the participants’ performance and brain activity. Moreover, we observed the reported distinction between automatic and controlled memory processes which play an integral part in both ageing and cognitive impairments.

Recent developments in Brain-Computer Interfaces (BCI)—technologies to collect brain imaging data—allow recording of Electroencephalography (EEG) data outside of a laboratory setting by means of mobile EEG systems. Brain imaging has been pivotal in understanding the neurobiological correlates of human behavior in many complex disorders. This is also the case for tinnitus, a disorder that causes phantom noise sensations in the ears in absence of any sound source. As studies have shown that tinnitus is also influenced by complexities in non-auditory brain areas, mobile EEG can be a viable solution in better understanding the influencing factors causing tinnitus. Mobile EEG will become even more useful, if real-time EEG analysis in mobile experimental environments is enabled, e.g., as an immediate feedback to physicians and patients or in undeveloped areas where a laboratory setup is unfeasible. The volume and complexity of brain imaging data have made preprocessing a pertinent step in the process of analysis, e.g., for data cleaning and artifact removal. We introduce the first smartphone-based preprocessing pipeline for real-time EEG analysis. More specifically, we present a mobile app with a rudimentary EEG preprocessing pipeline and evaluate the app and its resource consumption underpinning the feasibility of smartphones for EEG preprocessing. Our proposed approach will allow researchers to collect brain imaging data of tinnitus and other patients in real-world environments and everyday situations, thereby collecting evidence for previously unknown facts about tinnitus and other conditions.

Shift workers who are essential contributors to our society, face high risks of poor health and wellbeing. To help with their problems, we collected and analyzed physiological and behavioral wearable sensor data from shift working nurses and doctors, as well as their behavioral questionnaire data and their self-reported daily health and wellbeing labels, including alertness, happiness, energy, health, and stress. We found the similarities and differences between the responses of nurses and doctors. According to the differences in self-reported health and wellbeing labels between nurses and doctors, and the correlations among their labels, we proposed a job-role based multitask and multilabel deep learning model, where we modeled physiological and behavioral data for nurses and doctors simultaneously to predict participants’ next day’s multidimensional self-reported health and wellbeing status. Our model showed significantly better performances than baseline models and previous state-of-the-art models in the evaluations of binary/3-class classification and regression prediction tasks. We also found features related to heart rate, sleep, and work shift contributed to shift workers’ health and wellbeing.

The ability to accurately and automatically recognize and count the repetitions of exercises using a single sensor is essential for technology-assisted exercise-based rehabilitation. In this paper, we present a single deep learning architecture to undertake both of these tasks based on multi-channel time-series data. The models are constructed and tested using the INSIGHT-LME [1] exercise dataset which consists of ten local muscular endurance (LME) exercises. For exercise recognition, we achieved an overall F1-score measure of 96% and for repetition counting, we were correct within an error of ±1 repetitions in 88% of the observed exercise sets. To the best of our knowledge, our approach of using the same deep learning model for both tasks using raw time-series sensor data information is novel.

The advances of sensing and computing technologies pave the way to develop novel applications and services for wearable devices. For example, wearable devices measure heart rate, which accurately reflects the intensity of physical exercise. Therefore, heart rate prediction from wearable devices benefits users with optimization of the training process. Conventionally, Cloud collects user data from wearable devices and conducts inference. However, this paradigm introduces significant privacy concerns. Federated learning is an emerging paradigm that enhances user privacy by remaining the majority of personal data on users’ devices. In this paper, we propose a statistically sound, Bayesian inference federated learning for heart rate prediction with autoregression with exogenous variable (ARX) model. The proposed privacy-preserving method achieves accurate and robust heart rate prediction. To validate our method, we conduct extensive experiments with real-world outdoor running exercise data collected from wearable devices.

Proprioception is fundamental for maintaining balance and moving-hence for daily living. As proprioception deficits may occur with aging, neurological and musculoskeletal (especially cervical) conditions, assessment of proprioception can be relevant for a very large cohort of individuals.

We designed a web page that allows measuring the neck joint position sense while sitting in front of a standard webcam. The web page tracks the subjects’ head movement and instructs them on how to perform a head repositioning accuracy protocol. We performed a test retest analysis of this tool in order to assess its feasibility and reliability. Eleven healthy subjects participated in two sessions over consecutive days, at their homes. We calculated average errors across four directions Bland-Altman level of agreement between the measurements on the two sessions.

All participants could complete the test in approximately six minutes. The average absolute error did not differ between the two sessions, showing close to zero bias and a 95% limit of agreement of 1.676°. These values changed significantly across directions, suggesting that the performance of the head tracking software for neck flexion movements may be limited.

By comparing our results with normative values, we suggest that the narrow limit of agreement we observed makes the web page potentially capable of distinguishing healthy subjects from subjects with proprioceptive deficit in the neck joint.

This paper describes the process of developing the technical infrastructure of the Health Telescope: an interventional panel study designed to measure the long term effects of eHealth usage. We describe the design and implementation of both the Health Telescope application—an Android application that allows us to interact with participants and obtain measurements—and the researcher authoring client—a web-based application that allows us to flexibly submit experience sampling tasks to participants. This paper serves as a blueprint for those wanting to study long-term behavioral change in the wild. The paper furthermore describes a pilot study that was conducted to evaluate the research software. We conclude with design guidelines aimed at those aiming to undertake a similar endeavor that are vital when developing similar software; this paper aims to highlight both the importance and challenges of measuring the effects of eHealth applications longitudinally.

Mobile devices are becoming more pervasive in the monitoring of individuals’ health as device functionalities increase as does overall device prevalence in daily life. Therefore, it is necessary that these devices and their interactions are usable by individuals with diverse abilities and conditions. This paper assesses the usability of a neurocognitive assessment application by individuals with Parkinson’s Disease (PD) and proposes a design that focuses on the user interface, specifically on testing instructions, layouts, and subsequent user interactions. Further, we investigate potential benefits of cognitive interference (e.g., the addition of outside stimuli that intrude on task-related activity) on a user’s task performance. Understanding the population’s usability requirements and their performance on configured tasks allows for the formation of usable and objective neurocognitive assessments.

The delivery of a patient, to the operating theatre, in every hospital, consists of several heterogeneous departments working synchronously via communicating and sharing information, in relation to the current state of a patient’s care, as they travel through the surgical care-path way. The surgical care-pathway typically starts at admissions and finishes as the patient is leaving recovery. The problem being, as a patient navigates the care-pathway, there are numerous risk factors in the forms of technical, environmental and human that can influence a delay in the delivery of care. This paper will discuss these risk factors and highlight different approaches taken by several authors to address such issues. Additionally, a software application will be discussed that has being developed by the author that uses portable mobile devices, to address similar issues, for a private health care provider in the south of Ireland. The results of implementing the new solution show a potential decrease in patient throughput time and an overall increase of task visibility, across the surgical care-pathway.

The COVID-19 pandemic in 2020 has resulted in increased fatality rates across the world and has stretched the resources in healthcare facilities. There have been several proposed efforts to contain the spread of the virus among humans. Some of these efforts involve appropriate social distancing in public places, monitoring and tracking temperature at the point of access, etc. In order for us to get back to the “new normal", there is a need for automated and efficient human contact tracing that would be non-intrusive and effective in containing the spread of the virus. In this paper, we have developed “Connect", a Blockchain and Self-Sovereign Identity (SSI) based digital contact tracing platform. “Connect" will provide an automated mechanism to notify people in their immediate proximity of an occurrence of a positive case and would reduce the rate at which the infection could spread. The platform’s self-sovereign identity capability will ensure no attribution to a user and the user will be empowered to share information. The ability to notify in a privacy-preserving fashion would provide businesses to put in place dynamic and localized data-driven mitigation response. “Connect’s" SSI based identity wallet platform encodes user’s digital identities and activity trace data on a permissioned blockchain platform and verified using SSI proofs. The user activities will provide information, such as places travelled, travel and dispatch updates from the airport etc. The activity trace records can be leveraged to identify suspected patients and notify the local community in real-time. Simulation results demonstrate transaction scalability and demonstrate the effectiveness of “Connect" in realizing data immutability and traceability.

According to the United States’ Center for Disease Control and Prevention (CDC) between 39 and 56 million people in the US suffered from Influenza Like Illnesses (ILI) in the 2019-20 flue season. From which, 410 to 740 thousand were hospitalized and 24 to 62 thousand succumbed to the disease. Therefore, the existence of an early warning mechanism that can alert pharmaceuticals, healthcare providers, and governments to the trends of the influenza season well in advance, would serve as a significant step in helping combat this communicable disease and reduce mortality from it.

As reported in the [ACM Special Interest Group in Computers and Society (SIGCAS) 2020 Computers and Sustainable Societies (COMPASS)], [IEEE Technology and Engineering Management Society (TEMS) 2020 International Conference on Artificial Intelligence for Good (AI4G)], and [IEEE Global Humanitarian Technology Conference (GHTC) 2020] Long Short-Term Memory (LSTM) neural networks are utilized by Santa Clara University’s EPIC (Ethical, Pragmatic, and Intelligent Computing) and BioInnovation & Design laboratories for continued research and development of an eVision (Epidemic Vision) machine learning tool to predict the trend of influenza cases throughout the flu season.

There we reported eVision’s success in making 3, 7, and 14 weeks in advance predictions for the 2018–2019 United States flu season with 88.11%, 88%, and 74.18% accuracy respectively and delineated future steps of expanding eVision’s granularity by 1) adding state level predictions in order to enhance national predictions and 2) utilizing metropolitan area keyword trends to improve both state level and national predictions. This resulted in the improvement of the model’s accuracy to 90.38%, 91.43%, and 81.74% for 3, 7, and 14 weeks in advance predictions respectively. This paper is to report on the methodology of obtaining these improved results.

Neural Networks are powerful classifiers. However, they are black boxes and do not provide explicit explanations for their decisions. For many applications, particularly in health care, explanations are essential for building trust in the model. In the field of computer vision, a multitude of explainability methods have been developed to analyze Neural Networks by explaining what they have learned during training and what factors influence their decisions. This work provides an overview of these explanation methods in form of a taxonomy. We adapt and benchmark the different methods to time series data. Further, we introduce quantitative explanation metrics that enable us to build an objective benchmarking framework with which we extensively rate and compare explainability methods. As a result, we show that the Grad-CAM++ algorithm outperforms all other methods. Finally, we identify the limits of existing explanation methods for specific datasets, with feature values close to zero.

The classification of skin lesion images is known to be biased by artifacts of the surrounding skin, but it is still not clear to what extent masking out healthy skin pixels influences classification performances, and why. To better understand this phenomenon, we apply different strategies of image masking (rectangular masks, circular masks, full masking, and image cropping) to three datasets of skin lesion images (ISIC2016, ISIC2018, and MedNode). We train CNN-based classifiers, provide performance metrics through a 10-fold cross-validation, and analyse the behaviour of Grad-CAM saliency maps through an automated visual inspection. Our experiments show that cropping is the best strategy to maintain classification performance and to significantly reduce training times as well. Our analysis through visual inspection shows that CNNs have the tendency to focus on pixels of healthy skin when no malignant features can be identified. This suggests that CNNs have the tendency of “eagerly” looking for pixel areas to justify a classification choice, potentially leading to biased discriminators. To mitigate this effect, and to standardize image preprocessing, we suggest to crop images during dataset construction or before the learning step.

Many research articles used Machine Learning (ML) for early detection of Alzheimer’s Disease (AD) especially based on Magnetic Resonance Imaging (MRI). Most ML algorithms depend on a large number of hyperparameters. Those hyperparameters have a strong influence on the model performance and thus choosing good hyperparameters is important in ML. In this article, Bayesian Optimization (BO) was used to time-efficiently find good hyperparameters for Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) models, which are based on four and seven hyperparameters and promise good classification results. Those models are applied to distinguish if mild cognitive impaired (MCI) subjects from the Alzheimer’s disease neuroimaging initiative (ADNI) dataset will prospectively convert to AD. The results showed comparable cross-validation (CV) classification accuracies for models trained using BO and grid-search, whereas BO has been less time-consuming. The initial combinations for BO were set using Latin Hypercube Design (LHD) and via Random Initialization (RI). Furthermore, many models trained using BO achieved better classification results for the independent test dataset than the model based on the grid-search. The best model achieved an accuracy of 73.43% for the independent test dataset. This model was an XGBoost model trained with BO and RI.

Coronary Artery heart Disease (CAD) is the leading cause of mortality in the world. It is a complex and multifactorial disease resulting in several acute coronary syndromes and lead to death. In healthcare, an accurate clinical decision support system (CDSS) for CAD prediction has become increasingly important for making granted decisions at premature stage. Intensive research has been conducted on improving classification performance using machine learning techniques and metaheuristics algorithms. But most of these studies introduced the “classic risk factors” for CAD diagnosis i.e., demographic and clinical data. In this study, we present a novel CDSS based on ensemble learning for CAD prediction and we emphasize on adding other medical markers i.e., therapy data, some genetic polymorphisms along with classical factors. The new framework exploits the potential of three base classifiers including Support Vector Machines, Naïve Bayes and Decision Tree C4.5 to improve the prediction performance. Six experimental data used to build the proposed framework: the first one is collected from a Tunisian biotechnology center and the five other datasets from the University of California at Irvine repository. The analysis of the results shows that the proposed CDSS has the highest rate on classification accuracy, precision, recall and F1-measure when compared with CSGA Bagging and Adaptive boosting on the different datasets and proves that some medications and genetic polymorphisms such as Antivitamin K, Dose Beta Blocker, Proton pump inhibitor, CYP2C19*17, Clopidogrel active metabolite have an impact in CAD diagnosis.

The use of contrast agents in CT angiography examinations holds a potential health risk for the patient. Despite this, often unintentionally an excessive contrast agent dose is administered. Our goal is to provide a support system for the medical practitioner that advises to adjust an individually adapted dose. We propose a comparison between different means of feature encoding techniques to gain a higher accuracy when recommending the dose adjustment. We apply advanced deep learning approaches and standard methods like principle component analysis to encode high dimensional parameter vectors in a low dimensional feature space. Our experiments showed that features encoded by a regression neural network provided the best results. Especially with a focus on the 90% precision for the “excessive dose” class meaning that if our system classified a case as “excessive dose” the ground truth is most likely accordingly. With that in mind a recommendation for a lower dose could be administered without the risk of insufficient contrast and therefore a repetition of the CT angiography examination. In conclusion we showed that Deep-Learning-based feature encoding on clinical parameters is advantageous for our aim to prevent excessive contrast agent doses.

The severity of COVID-19 varies dramatically, ranging from asymptomatic infection to severe respiratory failure and death. Currently, few prognostic markers for disease outcomes exist, impairing patient triaging and treatment. Here, we train feed-forward neural networks on electronic health records of 819 confirmed SARS-CoV-2 positive patients admitted to a two-site NHS Trust hospital in London, England. To allow early risk assessment, the models ingest data collected in the emergency department (ED) to predict subsequent admission to intensive care, need for mechanical ventilation and in-hospital mortality. We apply univariate selection and recursive feature elimination to find the minimal subset of clinical variables needed for accurate prediction. Our models achieve AUC-ROC scores of 0.78 to 0.87, outperforming standard clinical risk scores. This accuracy is reached with as few as 13% of clinical variables routinely collected within the ED, which increases the practical applicability of such algorithms. Hence, state-of-the-art neural networks can predict severe COVID-19 accurately and early from a small subset of clinical variables.

Hypertension is one of the most common health conditions in modern society. Accurate blood pressure monitoring in free-living conditions is important for the precise diagnosis and management of hypertension. In tandem with the advances in wearable and ubiquitous technologies, a medical-grade wearable blood pressure monitor–Omron HeartGuide^TM wristwatch–has recently entered the consumer market. It uses the same mechanism as the upper arm blood pressure monitors and has been calibrated in laboratory settings. Nevertheless, its accuracy “in the wild” has not been investigated. This study aims to investigate the accuracy of the HeartGuide^TM against a medical-grade upper arm blood pressure monitor HEM-1022 in free-living environments. Analysis results suggest that the HeartGuide^TM significantly underestimated systolic pressure and diastolic pressure by an average of 16 mmHg and 6 mmHg respectively. Lower discrepancy between the two devices on diastolic pressure was observed when diastolic pressure increased. In addition, the two devices agreed well on heart rate readings. We also found that device accuracy was related to systolic pressure, heart rate, body temperature and ambient temperature, but was not related salivary cortisol level, diastolic pressure, ambient humidity and air pressure.

We present a prototype whereby we enabled a humanoid robot to be used to assist mental health patients and their families. Our approach removes the need for Cloud-based automatic speech recognition systems to address healthcare privacy expectations. Furthermore, we describe how the robot could be used in a mental health facility by giving directions from patient selection to metrics for evaluation. Our overarching goal is to make the robot interaction as natural as possible to the point where the robot can develop artificial empathy for the human companion through the interpretation of vocals and facial expressions to infer emotions.