Deep learning for detecting visually impaired cataracts using fundus images

Purpose: To develop a visual function-based deep learning system (DLS) using fundus images to screen for visually impaired cataracts. Materials and methods: A total of 8,395 (5,245 subjects) with corresponding function parameters collected from three clinical centers were used evaluate DLS classifying non-cataracts, mild cataracts, Three algorithms (DenseNet121, Inception V3, ResNet50) leveraged train models obtain the best one system. The performance was evaluated area under receiver operating characteristic curve (AUC), sensitivity, specificity. Results: AUC algorithm (DenseNet121) on internal test dataset two external datasets 0.998 (95% CI, 0.996–0.999) 0.999 0.998–1.000),0.938 0.924–0.951) 0.966 0.946–0.983) 0.937 0.918–0.953) 0.977 0.962–0.989), respectively. In comparison between cataract specialists, better observed in detecting cataracts ( p < 0.05). Conclusion: Our study shows potential function-focused screening tool identify images, enabling timely patient referral tertiary eye hospitals.