MonteRET: Region-Aware AI Agent Boosts Chest CT Report Recall
New retrieval-enhanced framework cuts omitted findings in chest CT reports by improving recall.
Automated chest CT report generation has long struggled with clinically faithful reporting that captures both whole-volume context and localized anatomical details. A team of 11 researchers from multiple institutions (including Weill Cornell, University of Texas, and others) introduced MonteRET, a novel region-aware retrieval-enhanced framework that addresses this gap. MonteRET integrates global CT features with region-level anatomical representations, retrieves clinically relevant knowledge using predicted medical conditions and region-level vision-language alignment, and refines initial reports through a knowledge-guided report rewriting agent.
The model was trained on the public RadGenome-ChestCT dataset comprising 24,128 CT scans. Evaluation on a test set of 1,564 scans and an external cohort of 82 CT scans from NewYork-Presbyterian/Weill Cornell Medical Center showed that MonteRET outperformed a matched baseline and several state-of-the-art methods. Gains were most pronounced for recall, indicating fewer omitted findings. Human expert evaluation by radiology residents also favored MonteRET, validating its clinical utility. The work is currently available on arXiv and represents a significant step toward reliable AI-assisted radiology reporting.
- MonteRET trained on 24,128 CT scans and tested on 1,564 internal + 82 external scans from a major medical center.
- Achieved notable improvements in recall, meaning fewer clinically relevant findings were missed compared to baselines.
- Radiology residents preferred MonteRET's reports in human evaluation, confirming real-world clinical value.
Why It Matters
Fewer missed findings in chest CT reports directly improves patient outcomes and reduces radiologist workload.