LLM-Based Late Fusion for Heterogeneous Echocardiography AI Outputs
October 30, 2026
A supervised LLM-based late-fusion module aggregates view-specific echocardiography findings and measurements directly in language space. In a study of 859 examinations, the method performed comparably to best-in-class reference strategies and outperformed general-purpose LLMs without task-specific training.
HOW THIS AFFECTS YOU
●
researcherThe method provides a framework for fusing heterogeneous model outputs via language-space aggregation.
●
healthThis could improve diagnostic accuracy by combining diverse ultrasound model outputs into a unified linguistic report.