Explore a cutting-edge approach to predicting patient diagnoses in critical care units using graph-based machine learning techniques. Delve into a presentation that showcases how de-identified healthcare data from over 40,000 patients is utilized to create evolving graphs representing patient states. Learn about the four key predictive features considered: fluid intake, fluid output, lab test results, and prescribed medications. Discover how these graphs are fed into a Recurrent Neural Network to predict the probability of top-K diagnoses at each step. Gain insights into the significant preliminary results achieved by leveraging graph-based ML techniques to convey relational information from previous patient admissions. Understand the potential of combining graphs and advanced ML techniques in pushing automated healthcare to new frontiers.
Overview
Syllabus
[VDZ19] Graph-based ML for Automated Health Care Services by J. Kindelsberger, A. Fritzen, R. Patra
Taught by
Devoxx