MedVersa Delivers Unified Multimodal Radiology Assistant

Researchers published in NEJM AI introduce MedVersa, a multimodal foundation model trained on over 29 million medical imaging instances from 91 datasets to perform detection, segmentation, and report generation. In blinded 2026 evaluations, MedVersa produced clinically comparable chest X-ray reports to human radiologists in 64% of cases (91% for normal scans), outperformed specialist models on several metrics, and reduced report-writing time and urgent discrepancies.