Classification of Bacterial Keratitis Activity with Patch-Based Deep Learning Using Three Anterior Segment Images

Bacterial keratitis is one of the common corneal diseases. Without timely and appropriate treatment, it can lead to complications such as vision reduction and perforation may occur, and in severe cases, it may even lead to blindness. In this study, aim to develop an artificial intelligence model for activity classification in bacterial keratitis using three types of anterior segment images: broad, slit, and scatter. By applying the patch technique, the highest AUROC of the model trained from the original was improved from 0.802 to 0.897. Performance experiments based on image combinations demonstrated that the model using only slit images showed the best results.