【Team】 AI Computing Team
【Date】2025/September/18(Thursday) 9:00-10:00(JST)
【Speaker】Talk by Yafei Ou, Institute of Science Tokyo
Title: Constraint-Aware Medical Imaging AI: Hardware-Friendly Rendering, Synthetic Priors, and Quantitative Longitudinal Monitoring
Abstract:
Clinical imaging AI rarely operates under ideal conditions, often facing tight computational budgets and scarce, heterogeneous data. This talk outlines a framework that treats these realities as design inputs and focuses on three aspects: (i) hardware-friendly rendering that preserves fidelity and controllability on resource-limited devices; (ii) synthetic, anatomy-aware pretraining that reduces annotation demand and stabilizes learning under distribution shift; and (iii) registration-driven longitudinal quantification that turns baseline and follow-up differences into reliable indicators. Together, these elements yield a small-data-ready, deployment-oriented pipeline that generalizes across centers, supports interpretable decisions, and is practical for on-device inference.
