Deep Learning for Detection of Corneal Perforation on Anterior Segment Optical Coherence Tomography in Microbial Keratitis

Purpose: To develop and evaluate deep learning models for automated detection of corneal perforation in microbial keratitis using anterior segment optical coherence tomography (ASOCT) images. Methods: We enrolled 150 patients with microbiologically confirmed keratitis. Contralateral healthy eyes served as controls. Four convolutional neural network models using ResNet architecture were trained and evaluated using ASOCT images to classify the presence or absence of corneal perforation at the eye level. Ground truth labels for perforation were established following consensus grading by two masked ophthalmologist graders. Models differed in inclusion of healthy controls and masking of non-corneal anterior segment anatomy. Results: The best-performing model (Model 1), which included healthy controls and randomly applied masking of the inferior image portion during training, achieved an AUC of 0.965 (95% CI, 0.911-0.995), sensitivity of 84.0% (95% CI, 70.0%-97.1%), and specificity of 97.8% (95% CI, 96.1%-99.3%) for detection of corneal perforation. Models including healthy controls outperformed those without, and lens masking improved discrimination. Conclusions: Deep learning models achieved high diagnostic accuracy for detecting corneal perforation on ASOCT imaging in eyes with microbial keratitis. These findings support the potential role of automated ASOCT analysis as a clinical decision support tool for identifying this vision-threatening complication.