WPI - Computer Science Department, PhD Proposal Defense, Atifa Sarwar " Machine Learning Models for Passive Assessment of Covid-19 from Smart Wearable Data"
WPI – Computer Science Department
Friday, March 17th, 2023
Time: 12:30 pm – 1:30 pm
Location: Fuller Labs 141
Advisor: Prof. Emmanuel Agu – WPI - Computer Science Department
Prof. Erin Solovey – WPI - Computer Science
Prof. Yanhua Li – WPI - Computer Science
External Member: Prof. Bashima Islam – WPI- Electrical and Computer Engineering Department
Covid-19, a recently discovered Influenza Like Illness (ILI), is an infectious disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Covid-19 vaccinations were introduced in early 2021, significantly reducing adverse outcomes including non-Intensive Care Unit (ICU) hospitalizations and deaths by 63.5% and 69.5% respectively in United States. However, due to mutating nature of the virus, Covid-19 is still a concern around the globe. Passive, continuous monitoring can facilitate early detection and timely interventions to reduce disease spread. In this dissertation we focus on heart rate, and steps data collected passively using a consumer-grade wearables (Fitbit and Apple Watch) to capture COVID-19 without requiring any prior history or human-reported symptoms. In pursuit of our dissertation goal, we have already published two papers: 1) Passive COVID-19 Assessment using Machine Learning on Physiological and Activity Data from Low End Wearables: A preliminary study to investigate whether changes in physiological (heart rate), activity and sleep variables captured by low-end wearables can be utilized to reliably detect COVID-19 using machine learning. II) CovidRhythm: Most living organisms have an internal biological clock that regulate their physiological functions, including performance, sleep, rest-activity cycles, and mood. Recent studies have found that SARS-CoV-2, disrupt the body clock to enhance their replication. Our second proposed method utilize disruptions in biobehavioral rhythms, including physiological (heart rate), and rest-activity rhythms (steps) to predict Covid-19. We propose CovidRhythm, a novel Gated Recurrent Unit (GRU) Network with Multi-Head Self-Attention (MHSA) that combines sensor and rhythmic features extracted from heart rate and activity (steps) data gathered passively using consumer-grade smart wearables.
To complete this dissertation, three more innovative solutions to predict Covid-19 from physiological symptoms are proposed, including:
I) A Continual Learning model for improving diagnoses by mitigating inter and intra-subject variability in the physiological symptoms (heart rate and steps.
II) A Graph Neural Network that leverages intra-patient similarities, so that only the relevant data points (patients) contribute to the prediction of a given sample, dampening the impact of irrelevant samples.
III) A Reinforcement Learning based agent that aim to find the optimal length of the input data that can predict Covid-19 early in the incubation period (5 days before symptom onset) with as few measurements as possible.