Date of Award
11-14-2024
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
The Internet of Medical Things (IoMT), which integrates Internet of Things technologies into healthcare, has become a powerful tool for enhancing health monitoring and predicting clinical outcomes. By leveraging wearable devices, IoMT facilitates continuous, cost-effective, and convenient tracking of patients' health conditions over time. This dissertation applies data-driven methods to address critical clinical challenges involving wearable devices. Specifically, it focuses on three significant clinical problems: (1) indoor contact tracing for healthcare workers using Bluetooth Low Energy (BLE) beacons, (2) predicting surgical outcomes using wearable data, and (3) developing robust models for surgical outcome prediction that account for patient variability using wearable data. Despite advancements in these areas, challenges such as incomplete and noisy data from wearable devices have been largely overlooked. Additionally, patient variability, a common issue in clinical studies, has not been thoroughly investigated. To address these challenges, we propose novel machine learning approaches designed to handle noisy and incomplete wearable data, as well as account for patient variability. Specifically, this thesis research makes three major contributions. 1) For contact tracing, a set of contact tracing methods is tailored specifically to the heterogeneous hospital environment and clinicians’ working patterns. 2) For the surgical outcome prediction, an end-to-end feature engineering approach is developed to extract clinically meaningful features from step count, heart rate, and sleep data recorded by wearable wristbands. 3) Furthermore, we introduce a Mixture of Experts model, integrated with a novel diversity loss function, to fuse wearable features with traditional static clinical features, accounting for patient variability and ultimately producing a more robust model for surgical outcome prediction. Our findings suggest that addressing incomplete and noisy wearable data, along with patient variability, leads to more accurate and robust clinical predictive models, thereby enhancing the reliability of healthcare decision-making.
Language
English (en)
Chair
Chenyang Lu
Committee Members
Chenguang Wang; Chet Hammill; Maria Cristina Vazquez Guillamet; Thomas Kannampallil; Yevgeniy Vorobeychik