Building trusted AI for precision medicine: from multimodal clinical data to federated learning and synthetic cohorts

Abstract

Artificial intelligence is changing how clinical data are used to understand disease, predict outcomes, and support more personalized patient care. In this lecture, Dr. Maros will present examples from his research on real-world clinical data, including medical imaging, text, and molecular data, with applications in stroke, brain tumors, dementia, and predictive disease modeling.

The lecture will focus on the opportunities and challenges of building trusted AI for clinical use, including model robustness, calibration, data quality, shift, and the evaluation of both self-developed and commercial AI tools. It will also address how federated learning, predictive forecasting, and synthetic cohort generation can support privacy-preserving, scalable, and clinically meaningful precision medicine.