Unified Diagnosis Framework for Automated Nuclear Cataract Grading Based on Smartphone Slit-Lamp Images

Abstract

Cataract constitutes half of the blindness cases worldwide; hence, detecting and treating cataracts in a timely manner are effective strategies for blindness prevention. Recently, methods of detecting cataracts through deep learning are flourishing; however, the task of improving the grading mechanism is still the priority in the research field. This study evaluates the classification capability of the automated nuclear cataract detection algorithm using ocular images captured by smartphone-based slit-lamp. The task of the algorithm is to automatically detect cataract severity in terms of the photometric appearance of the nuclear region of the crystalline lens of the eyes. The nuclear region of the ocular lens was localized by YOLOv3. Subsequently, the combination of a deep learning network, ShuffleNet, and a support vector machine (SVM) classifier was used to grade cataract severity, evaluating the gray conjugate features of the nuclear region. Using the trained algorithm, 819 anterior ocular images captured by smartphone-based slit-lamp were utilized to evaluate the algorithm's performance. The accuracy was 93.5% with Kappa of 95.4% and F1 of 92.3%. The AUC was 0.9198. The proposed validation method could evaluate a cataract severity in 29 ms and the entire classification process in less than 1s. This study can improve the accuracy of the examination, reduce misdiagnosis rate and the difficulty of the doctor's examination. The addition of scoring system can improve the quality of pictures obtained by non-ophthalmologists. The method is especially suitable for cataract screening in the underdeveloped areas or areas which are in shortage of ophthalmic resources. It can also improve the accessibility of ophthalmic medical treatment.

Keywords

• Images captured by smartphone-based slit-lamp
• automated cataract detection
• grade cataract
• deep learning