AI Tool CoRaX Integrates Eye Gaze and Reports to Reduce Chest X-Ray Errors
July 17, 2025
Researchers introduce CoRaX, an AI system that uses eye gaze and radiology report data to address perceptual errors in chest x-ray readings.
Key Details
- CoRaX combines eye gaze data, chest x-ray images, and radiology reports for collaborative error detection.
- Developed by the University of Houston, the system acts as a 'virtual second reader' and provides targeted diagnostic recommendations.
- Evaluation on simulated datasets showed CoRaX corrected 21.3% and 34.6% of perceptual errors, depending on the scenario.
- Diagnostic aid was provided in 85.7% and 78.4% of interactions in two different datasets.
- The system performed particularly well for missed cases of cardiomegaly.
- Future directions include real-world validation and possible integration of more advanced classifiers.
Why It Matters
Addressing perceptual errors is crucial to radiology safety and accuracy. CoRaX's collaborative AI approach could significantly enhance radiologists' performance, setting the stage for future clinical implementation and error-resistant diagnostic workflows.