The ReXGrounding Challenge, held at MICCAI 2026, aimed to improve AI's ability to segment findings in 3D chest CT scans, a task where current models perform poorly. The challenge utilized the ReXGroundingCT dataset with 2,992 training and 200 validation scans, evaluating 760 submissions from 48 teams. The winning method achieved a Dice score of 0.375 and a hit rate of 0.77, comparable to inter-radiologist agreement. However, challenges remain in segmenting non-focal findings, multiple instances of abnormalities, and accurately localizing segmented findings. AI
IMPACT Advances in grounding radiology reports in medical scans could lead to more verifiable AI-generated reports and improved patient monitoring.
RANK_REASON The item describes the results of a challenge focused on a specific research task in medical imaging AI, including performance metrics and open problems. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →