Multi-Modal AI Framework for Early Detection of Pancreatic Cancer

Prevention & Diagnostic Tools

Ludmil Alexandrov, PhD (UC San Diego)

Diane Simeone, MD (UC San Diego)

Adam Yala, PhD (UC Berkely)

Karandeep Singh, MD (UC San Diego)

Pancreatic ductal adenocarcinoma (PDAC) is among the deadliest cancers, with >80% diagnosed at advanced stages and <15% 5-year survival for locally advanced disease. Early detection markedly improves outcomes, yet current risk models miss most high-risk individuals. We hypothesize that integrating clinical, imaging, lifestyle, and genomic data through multi-modal artificial intelligence (AI) will enable accurate early PDAC risk prediction. We will develop: (1) an EHR/imaging model trained on >1M scans with matched clinical data from the UC Cancer Consortium, (2) a lifestyle/genomic model using Mutographs questionnaires and whole-genome sequencing, and (3) a unified predictive framework via ensemble learning. Validation in the independent prospective PRECEDE cohort (150 PDAC cases, 300 controls) will assess accuracy. The expected outcome is a clinically scalable tool for early PDAC detection, enabling targeted surveillance and earlier intervention to improve survival.