MMRad-22K: A Structured Multimodal Evidence Dataset for Chest X-ray Report Generation
Researchers have introduced MMRad-22K, a new dataset designed to improve chest X-ray (CXR) report generation. This dataset structures regional textual observations, anatomical coordinates, and image evidence into multimodal units. Experiments show that using this structured multimodal evidence with a unified LVLM backbone leads to better performance on language and clinical metrics compared to text-only or bounding box-based evidence. AI
IMPACT This dataset could improve the accuracy and clinical relevance of AI-generated chest X-ray reports.