Senior Honors Thesis Presentation: "Using Visual Field Data for Assessing Glaucoma Progression: a Comparative Analysis of Survival Prediction Models"

Abstract: The most distinctive feature of survival data is right censoring, meaning that the true survival time for a patient could be unknown if the individual is lost to follow up. The Cox Proportional Hazards Model (Cox, 1972) is the most widely used statistical method for analyzing survival data. However, this classical method has been challenged by modern machine learning algorithms such as Random Survival Forests (Ishwaran, 2008) and DeepSurv (Katzman, 2018) due to the exponential growth in computing power over the past decade. During the presentation, I will evaluate different survival methods on assessing Glaucoma Progression using the data from a study developed by WashU medical school. I will also debate about the prospect and restriction of using deep learning for risk prediction in clinical setting. This talk will not be theory-heavy, so anyone interested in the application of modern machine learning techniques in biomedical science is welcomed.

Hosts: Jimin Ding and Lei Liu