MACHINE LEARNING METHODS FOR MODERN BIOMEDICAL DATA: DEEP LEARNING AND REINFORCEMENT LEARNING IN PRECISION HEALTH

In recent years, there is an increasing need for analysis tools in healthcare setting that can process modern biomedical data such as biomedical signals and images as well as the need for precision medicine methodology to help guide the data-driven decision making process in healthcare context. Deep Learning has emerged as an important tool to process the aforementioned domain of data and Reinforcement Learning technique can leverage existing longitudinal datasets to guide interventions that improve clinical outcomes.We develop four methods in this context: an off-policy evaluation reinforcement learning analysis to inform health disparity in clinical outcomes, two novel deep learning models: one to process electrocardiogram signal for the task of cardiac rhythms detection and another to process ultrasound sweeps for gestational age estimation, and lastly a deep learning extension of the proportional odds model in survival analysis.Doctor of Philosoph