Deep Learning-Based Analysis of Ocular Anterior Segment Diseases from Patient-Self-Captured Smartphone Images

The goal of this study is to evaluate the Eye Home Clinic app (ver 1.0), which uses deep learning models to assess the quality of self-captured anterior segment images and detect anterior segment diseases using only the patient’s smartphone. Images undergo quality assessment based on the ‘DL-Image Eligibility’ model, and usable images are analyzed by the ‘DL-Diagnosis’ model to detect one of several anterior segment diseases. A dataset of 1006 images was used for training, and a dataset of 520 images was used for validation. The ‘DL-Image Eligibility’ model achieved an AUC of 0.87, with an accuracy of 0.75. The ‘DL-Diagnosis’ model had higher specificity (0.97) but lower sensitivity (0.29), with an AUC of 0.62. While the app shows potential for anterior segment telemedicine, improvements are needed in the DL model’s sensitivity for detecting abnormalities. Oversampling techniques, transfer learning, and dataset expansion should be considered to enhance the performance in future research. Based on data from users in over 100 countries, significant differences in photo quality among user groups were also identified. iOS users, younger users (21–40 years), and users reporting eye symptoms submitted more usable images. This study underscores the importance of user education and technological advancements to optimize smartphone-based ocular diagnostics.