Development and Validation of a Deep Learning Algorithm for the Detection of Corneal Opacities in Resource-Limited Settings

Purpose: This study aimed to develop and validate a convolutional neural network (CNN)-based deep learning algorithm for detecting corneal opacities from smartphone photographs. Methods: An EfficientNetB7 CNN was trained and validated using a dataset of 10,968 annotated corneal photographs from the Village-Integrated Eye Worker (VIEW) I trial in Nepal to perform image-based detection of corneal opacity. Model performance was evaluated using the area under the receiver operating characteristic curve, accuracy, sensitivity, and specificity. Results: The model achieved an area under the receiver operating characteristic curve of 0.97 (95% CI, 0.96–0.98), 93% accuracy (95% CI, 91%–94%), 91% sensitivity, and 93% specificity in the hold-out test set of 1109 photos. Conclusions: This deep learning computer vision model's high performance in detecting corneal opacities from smartphone photographs offers a scalable solution for teleophthalmology in resource-limited settings. Future research should validate the model across diverse populations and different smartphone devices to evaluate its impact on clinical outcomes.