Machine Learning for Personalized Cancer Care: Distinguished Scientist Director's Seminar Series
June 5, 2024
On June 5, 2024, Moores Cancer Center welcomed Adam Yala, PhD, who presented “Machine Learning for Personalized Cancer Care.” The talk was co-hosted by Moores Director, Diane Simeone, MD and Moores Associate Director for Basic Science, J. Silvio Gutkin, PhD. In a rapidly evolving healthcare landscape, especially with respect to technology and artificial intelligence, this topic is of particular importance for both Moores and UC San Diego Health, given the promise of machine learning for cancer detection.
Currently an Assistant Professor of Computational Precision Health, Electrical Engineering, and Computer Science at UC Berkeley and UCSF, Dr. Yala holds a BS, MEng, and PhD in Computer Science from MIT. His research targets the prediction of future cancer risk and the design of personalized screening policies, with an emphasis on determining the intersection between early detection and over screening.
Dr. Yala began his talk by addressing the current problem with cancer screening: As a one-size-fits-all approach, it often results in late detections and over-screening, which are exacerbated by device invariance that hinders effective risk modeling over time.
Concerning the challenges facing personalized cancer screening today, Dr. Yala said that “Too many people need more, and too many people need less.”
Because of the lack of personalized approaches available, resources often are misallocated. Dr. Yala asserts that to begin personalizing cancer care, early detection must be prioritized, while operationalizing the idea using machine learning techniques.
“The question is not looking at year 5; it is looking at year 0 to figure out if there are any signs that cancer will appear.”
After presenting the image-based risk model, Dr. Yala outlined the next steps for real-world implementation. He explained that current tools utilize only a tiny fraction of available data, which emphasizes the need for researchers to gain better control of data management. Dr. Yala stressed that optimizing screening decisions, based on imaging and understanding casual effects from real world trajectories, will enhance not only early detection, but also opportunities for personalization.
Against this backdrop, Dr. Yala argues that the ability to accurately access patient risk at any given time point and design screening regiments based on that risk is key. Increasing personalization in this way will enable researchers to receive consistent results across diverse populations.
“There is room to be much more ambitious in how we think about screening policies overall. It is not the way you build it that proves it; it is the way you test it that proves it.”
Dr. Yala advocates for the continued use of reusable and AI-adaptive randomized controlled trials to refine intervention strategies. This approach requires rethinking big picture guidelines across diseases to open new avenues for prevention and therapeutic development. To learn more about Dr. Yala’s research and publications, visit his website: https://www.adamyala.org.