Cheng Ding
BME PhD Proposal Presentation

Date: 2023-12-20
Time: 3PM-4:30PM
Location / Meeting Link: https://emory.zoom.us/j/93808230568

Committee Members:
Xiao Hu; Rishikesan Kamaleswaran; Eva Dyer; Vicki Hertzberg; Cynthia Rudin ; Ran Xiao


Title: Toward accurate health monitoring based large-scale Photoplethysmography signal from wearable devices

Abstract:
The thesis focuses on maximizing the use of large-scale photoplethysmography (PPG) signal datasets to enhance the precision of health monitoring. The first objective involves a semi-supervised method for automatically labeling PPG data using the imprecise alarms from bedside monitors. This process yielded 8.5 million 30-second PPG records. To address the label errors caused by the imprecise alarms, a robust learning strategy was developed, significantly improving the detection accuracy of atrial fibrillation (AF). The second objective delves into developing two foundational models to over 2.6 million hours of PPG data. These models include an artifact mitigation technique using the SimSiam architecture and a generative model inspired by GPT. The integration of these models is aimed at enhancing downstream health monitoring tasks, such as estimating heart rate and blood pressure, and detecting AF with greater accuracy.