Deep Learning-Based Analysis of Infrared Fundus Photography for Automated Diagnosis of Diabetic Retinopathy with Cataracts

Abstract

Purpose: To develop deep learning-based networks for the diagnosis of diabetic retinopathy (DR) with cataracts based on infrared fundus images. Setting: Shanghai General Hospital, Shanghai Eye Disease Prevention & Treatment Center Design: Development and evaluation of an artificial intelligence (AI) diagnostic method. Methods: For this study, we gathered a total of 10,665 infrared fundus images from 4,553 patients with diabetes to train and test our model. To achieve our goals of image quality assessment, left and right eye classification, DR diagnosis and grading, and segmentation of three DR lesions, we developed an end-to-end software using EfficientNet and UNet. We also evaluated the accuracy and performance of the software in comparison to human experts. Results: The model achieved an accuracy of 75.31% for left and right eye classification, 100% for DR grading and diagnosis tasks, and 73.67% for internal test set, with corresponding AUCs of 0.88, 1.00, and 0.89, respectively. For DR lesion segmentation, the AUCs of hemorrhagic, microangioma, and exudative lesions were 0.86, 0.66, and 0.84, respectively. In addition, a contrast test of human-machine film reading confirmed the software's high sensitivity (96.3%) and specificity (90.0%) and consistency with the manual film reading group (kappa=0.869, P<0.001). This easily deployable software can generate reports quickly and promote efficient DR screening with cataracts in clinical and community settings. Conclusion: AI-assisted software can perform automatic analysis of infrared fundus images and has substantial application value for the diagnosis of DR patients with cataracts.

Cataract surgery innovations

“…Development of deep learning–based networks using fundus images in the diagnosis of diabetic retinopathy in eyes with cataracts is seen to be of considerable application value in clinical diagnosis. [ 1 ] Comparison of automatic and manual segmentation deep transform learning platforms showed the automatic segmentation platform to stage cataracts more quickly, while the manual segmentation platform staged cataracts with increased accuracy. [ 2 ] Several AI-assisted automatic cataract grading program-based slit-lamp and retro illumination lens clinical photographs based on lens opacities classification system III that have been validated have shown accurate and precise detection and grading of various types of lens opalescence.…”