Transition-Aware Best-of-N Sampling for Chest X-ray Reports
June 22, 2026
This training-free sampling scheme improves longitudinal chest X-ray report generation by encoding the change between a prior and current exam. It uses a set-to-set distance to create a transition vector, scoring report candidates via cosine distance to ground-truth clinical changes.
HOW THIS AFFECTS YOU
●
builderYou can implement longitudinal context in report generators without retraining the base model.
●
healthThis improves the clinical utility of AI radiology assistants by focusing on relevant changes over time.