A tumor's microscopic appearance may contain clues about its underlying mutations, but extracting those clues reliably requires far more than training an AI algorithm.
Balint of SciAnts_Streams speaks with Dr. Lucas Liu of Fred Hutch Cancer Center about how machine learning is being developed to address some of precision oncology's most difficult problems: predicting cancer-associated mutations, estimating future disease risk, and integrating complex patient information into more individualized treatment strategies.
Dr. Liu explains how researchers train computational models to recognize relationships between cellular morphology, medical imaging, and molecular characteristics of cancer. The discussion examines efforts to predict breast cancer risk over several years and a prostate cancer research project investigating how a decade of clinical records might improve personalized predictions.
A central focus is the scientific challenge of building trustworthy AI. Preparing medical data, controlling for differences between hospitals, validating pathology images, and ensuring that algorithms generalize beyond their original training populations are essential steps. The interview also explores multimodal deep learning, in which separate models analyze different biological and clinical measurements before integrating their outputs.
The conversation examines the Cancer AI Alliance and federated learning, an approach that allows researchers at different institutions to collaborate on model development while keeping patient records within their respective institutions.
Importantly, Balint and Dr. Liu distinguish experimental predictive models from clinically validated diagnostic tools. An algorithm identifying a statistical association is not the same as establishing a biological mechanism, and promising performance in a research dataset does not automatically translate into reliable patient care.
This episode provides a detailed look at the computational methods, biological questions, and practical limitations shaping the development of AI-assisted cancer medicine.
Recorded on 9/16/2026
Fred Hutchinson Cancer Center Socials
Homepage: https://www.fredhutch.org/
Facebook: https://www.facebook.com/HutchinsonCenter
X: https://x.com/fredhutch
LinkedIn: https://www.linkedin.com/company/fredhutch/
TikTok: tiktok.com/@fredhutch
YouTube: https://www.youtube.com/fredhutch
Instagram: https://instagram.com/fredhutch