The optimal treatment strategy for volume resuscitation and vasopressor dosing to enhance outcomes of sepsis patients remains a subject of ongoing controversy. Our objective is to explore the clinical utility of machine-learning approaches in optimizing fluid and vasopressor-dosing strategy for improved sepsis outcomes in critical-care settings. Some recent research has focused on using reinforcement learning — specifically, deep Q-networks — to learn optimal policies for sepsis treatment with mortality as the reward function. Our project will focus on comparing policies learned using different reward functions in a deep reinforcement learning framework for improving the survival and end-organ functions of sepsis patients.

I’m interested in participating in SuperUROP because I wanted an opportunity to further enhance my research skills. I am continuing my research from last semester as a UROP scholar. I hope to gain an increased understanding of how the limitations of the medical field affects applied machine learning research. I’m excited about the opportunity to work with real patient data to solve a real medical problem.