Dr. Mayfield recently presented his groundbreaking research at the SRG Pillar meeting for the Center for Intelligent Imaging (ci2) at the University of California, San Francisco. His talk, “Foundation Models & Multi-Omic Methods: Revolutionizing Precision Medicine with Classical + Quantum AI,” highlighted how emerging AI systems are reshaping healthcare.

Foundation models—large-scale, pretrained AI systems—can be adapted to a wide range of downstream tasks through fine-tuning, much like large language models (LLMs). The success of LLMs has inspired similar strategies for medical data, with the aim of addressing challenges such as clinician burnout. Trained on vast, unlabeled datasets, these models demonstrate remarkable generalization across diverse clinical tasks.

Such advances are accelerating precision medicine, which depends on integrating diverse health data to achieve a holistic view of patient well-being. Dr. Mayfield emphasized that multi-omic data—spanning genomics, radiomics, and other “omics” layers—represents a critical subset of multimodal inputs. “By combining these data streams, we can uncover deeper insights into disease progression, patient response, and personalized treatment strategies,” he explained. “This integration forms the foundation of next-generation AI in medicine.”

While multi-omic approaches offer significant benefits—including improved prediction accuracy, early detection, targeted therapy selection, mechanistic insights, and novel biomarker discovery—Dr. Mayfield also underscored key challenges. Issues such as patient privacy, incomplete datasets, and limited interpretability remain important considerations.

He illustrated these opportunities and challenges through several case studies. In one project, he developed a semi-foundation model using 8,500 unlabeled chest CT scans, which enhanced classification of osteoporosis severity and fracture risk through self-supervised learning. In another study, a classical foundation model enabled early detection of HPV-positive squamous cell carcinoma.

Pushing further, Dr. Mayfield combined classical and quantum approaches to build a hybrid foundation model for glioblastoma outcome prediction. The classical component provided transformer-based embeddings of imaging and pathology data, while the quantum branch leveraged variational quantum circuits for genomic pattern recognition. This innovative framework earned him recognition in the 2024 NIH NCI/ODSS Biomedical Quantum Computing Challenge.

Source: https://intelligentimaging.ucsf.edu/news/revolutionizing-precision-medicine-ai