From AI Innovation to Clinical Impact: Closing the Gap for Reliable AI-based Disease Screening

PI: Irene Ong, Associate Professor of Obstetrics and Gynecology and Biostatistics and Medical Informatics

Co-PIs:

Maja Waldron, Assistant Professor of Statistics

Daniel Cho, Assistant Professor of Surgery

Fred Sala, Assistant Professor of Computer Sciences

Paula Voorheis, Assistant Professor of Pharmacy

Description: AI-based disease screening holds transformative promise, yet its translation into clinical practice has been slow. Three barriers are responsible: First, a Data Efficiency Gap: most real-world screening settings, from rare genetic conditions to ovarian cancer, involve small, fragmented, and multimodal datasets that standard machine learning pipelines cannot handle reliably. Second, a Clinical Integration Gap: even high-performing models fail to improve care when they do not fit clinical workflow, earn clinical trust, or adapt over time. Third, a Longitudinal Forecasting Gap: disease risk unfolds over time, but most models rely on single-timepoint snapshots rather than forecasting over longitudinal, multimodal electronic health record data — missing the trajectories that could flag risk before symptoms appear. These problems block AI impact across every screening context. The urgency is clear: AI capabilities are advancing rapidly; but without solving these translation barriers, the gap between algorithmic performance and clinical benefit will continue to widen.