Deep Learning-Based Cataract Detection and Grading from Slit-Lamp and Retro-Illumination Photographs: Model Development and Validation Study

Abstract

Backgroud: Accurate detection and grading of cataract is pivotal in providing personalized precision treatment strategies for preventing visual loss. We develop and validate an automated deep learning (DL)-based artificial intelligence (AI) platform for diagnosing and grading cataracts, using slit-lamp and retro-illumination lens photographs based on the Lens Opacities Classification System-Ⅲ (LOCS-Ⅲ). Methods: Slit-lamp and retro-illumination lens photographs were graded by two trained graders using LOCS-III. Image datasets were labeled and divided into training, validation, and test datasets. We trained and validated AI platforms with four key strategies in the AI domain: (1) region detection network for redundant information inside data, (2) data augmentation and transfer learning for the small dataset size problem, (3) generalized cross entropy loss for dataset bias, and (4) class balanced loss for class imbalance problems. The performance of the AI platform was reinforced with an ensemble of three AI algorithms (ResNet18, WideResNet50-2, and ResNext50). Finding: The AI platform showed robust diagnostic performance (AUC of 0.9992 [0.9986-0.9998], 0.9994 [0.9989-0.9998]; accuracy of 98.82% [97.7-99.9], 98.51% [97.4-99.6]) and grading prediction performance (AUC of 0.9567 [0.9501-0.9633], 0.9650 [0.9509-0.9792]; accuracy of 91.22% [89.4-93.0], 90.26% [88.6-91.9]) for nuclear opalescence (NO) and nuclear color (NC) using slit-lamp photographs. For cortical opacity (CO) and posterior subcapsular opacity (PSC), the system achieved high diagnostic performance (AUC of 0.9680 [0.9579-0.9781], 0.9465 [0.9348-0.9582]; accuracy of 96.21% [94.4-98.0], 92.17% [88.6-95.8]) and good grading prediction performance (AUC of 0.9044 [0.8958-0.9129], 0.9174 [0.9055-0.9295]; accuracy of 91.33% [89.7-93.0], 87.89% [85.6-90.2]) using retro-illumination images. Interpretation: Our DL-based AI platform successfully yielded accurate and precise detection and grading of NO and NC in seven-level classification and CO and PSC in six-level classification, overcoming the limitations of medical databases such as small training data or biased label distribution. Funding: This work was supported by the National Research Foundation of Korea grant funded by the Korean government's Ministry of Education (NRF-2021R1C1C1007795; Seoul, Korea), which was received by D.H.L. Declaration of Interest: The authors declare no conflict of interest Ethical Approval: The study was approved by the Samsung Medical Center Institutional Review Board (IRB file number 2020-08-035)