Midweek Keynote: HPC for Vascular Digital Twins

In this midweek keynote, Amanda Randles of Duke University, director of the Center for Computational and Digital Health Innovation, describes how high performance computing enables patient-specific vascular digital twins that could shift medicine from reactive care toward proactive, predictive monitoring. She frames a digital twin as a continuously evolving computational model driven by real sensor data, built from medical imaging (CT/MRI) that is segmented into 3D vascular geometry and fed into large-scale computational fluid dynamics using the Harvey flow solver. Randles traces three challenges: scale, time, and clinical usefulness. For scale, she reviews decades of CFD progress and introduces adaptive physics refinement (APR), a high-resolution window that explicitly models red blood cells and cancer-cell adhesion near a tracked cell while treating downstream flow as bulk fluid, dramatically cutting cost versus brute-force simulation on systems like Summit. For time, the longitudinal hemodynamic mapping framework breaks long streams of heartbeats into parallel, reusable chunks driven by wearable data (Fitbit, Apple Watch), reproducing weeks of wall shear stress exposure with under 1 percent error. For clinical use, she cites fractional flow reserve, pulmonary artery pressure matching, and machine-learning tools that predict treatment outcomes in real time. Randles emphasizes validated, interpretable AI rather than black-box models, and cautions that wearable signals alone are insufficient without physics-based, geometry-specific simulation.