AI Foundation Models & Molecular Biomarkers in Breast Cancer

In this episode of The AI Oncology Revolution, host Arturo Loaiza-Bonilla, MD, MSEd, FACP, sat down with Frederick Howard, MD, an assistant professor of Medicine at the University of Chicago, to explore the evolving intersection of artificial intelligence (AI), computational pathology, and precision breast oncology.

Howard shared his background in computer science and clinical research, offering a deep dive into how pathology foundation models are moving beyond routine image automation to identify novel molecular biomarkers, predict disease recurrence, and assess treatment sensitivity.

The conversation covered key clinical considerations including:

• Pathology Foundation Models: How AI models trained on millions of pathology slides are decoding complex tumor phenotypes and predicting molecular features like Oncotype DX scores.
• Evaluating Clinical Utility: Why high prognostic value alone is insufficient, and what level of evidence is required to prove true predictive value for chemotherapy benefit in HR-positive/HER2-negative disease.
• Model Validation & Pitfalls: Addressing computational challenges such as batch effects, shortcut learning, and the necessity of head-to-head comparisons against current standard-of-care genomic assays.
• Biomarker Standardization: The potential for AI-driven image analysis to resolve inter-observer discordance in nuanced settings like HER2-low and HER2-ultralow identification.
• The Multimodal Future: How integrating digital pathology, genomics, and large-scale electronic health record datasets, such as Epic Cosmos, will drive next-generation clinical decision support.