Researchers from the University of Missouri deploy Mask R-CNN for precise corneal segmentation followed by ResNet50 transfer learning to classify sulfur mustard–induced rabbit eye injuries into four severity grades. This automated pipeline reduces diagnostic variability and enhances translational potential for ocular chemical injury studies.
Key points
- Mask R-CNN segments corneal regions to isolate relevant injury areas from stereomicroscope images.
- ResNet50 transfer learning classifier reaches 87% training accuracy and 85%/83% test accuracies across independent datasets.
- Study uses 401 sulfur mustard–exposed rabbit corneal images with nested k-fold cross-validation to ensure model robustness.
Why it matters: This AI-driven grading system sets a new standard for consistent, rapid, and objective assessment of ocular chemical injuries, expediting preclinical research and therapeutic development.
Q&A
- What is Mask R-CNN segmentation?
- Why use transfer learning with ResNet50?
- How does objective AI grading benefit research?
- What do ROC-AUC and Hamming distance measure?
