Deep learning model for extensive smartphone-based diagnosis and triage of cataracts and multiple corneal diseases

Abstract

Aim: To develop an artificial intelligence (AI) algorithm that diagnoses cataracts/corneal diseases from multiple conditions using smartphone images.

Methods: This study included 6442 images that were captured using a slit-lamp microscope (6106 images) and smartphone (336 images). An AI algorithm was developed based on slit-lamp images to differentiate 36 major diseases (cataracts and corneal diseases) into 9 categories. To validate the AI model, smartphone images were used for the testing dataset. We evaluated AI performance that included sensitivity, specificity and receiver operating characteristic (ROC) curve for the diagnosis and triage of the diseases.

Results: The AI algorithm achieved an area under the ROC curve of 0.998 (95% CI, 0.992 to 0.999) for normal eyes, 0.986 (95% CI, 0.978 to 0.997) for infectious keratitis, 0.960 (95% CI, 0.925 to 0.994) for immunological keratitis, 0.987 (95% CI, 0.978 to 0.996) for cornea scars, 0.997 (95% CI, 0.992 to 1.000) for ocular surface tumours, 0.993 (95% CI, 0.984 to 1.000) for corneal deposits, 1.000 (95% CI, 1.000 to 1.000) for acute angle-closure glaucoma, 0.992 (95% CI, 0.985 to 0.999) for cataracts and 0.993 (95% CI, 0.985 to 1.000) for bullous keratopathy. The triage of referral suggestion using the smartphone images exhibited high performance, in which the sensitivity and specificity were 1.00 (95% CI, 0.478 to 1.00) and 1.00 (95% CI, 0.976 to 1.000) for 'urgent', 0.867 (95% CI, 0.683 to 0.962) and 1.00 (95% CI, 0.971 to 1.000) for 'semi-urgent', 0.853 (95% CI, 0.689 to 0.950) and 0.983 (95% CI, 0.942 to 0.998) for 'routine' and 1.00 (95% CI, 0.958 to 1.00) and 0.896 (95% CI, 0.797 to 0.957) for 'observation', respectively.

Conclusions: The AI system achieved promising performance in the diagnosis of cataracts and corneal diseases.

Keywords: Cornea; Ocular surface.

Figure 1. Selection process and sample size for training and testing sets. (A) Representative slit-lamp photographs of nine categories. (B) A total of 15 498 anterior segment photographs of 9 categories were captured using slit-lamp microscopy with diffuser light and 973 images were captured using an iPhone 13 Pro. IOL, intraocular lens.

Figure 2. Performance of deep learning algorithm to classify cataract/cornea diseases into nine categories. Receiver operating characteristic curves indicating performance of YOLO V.5 for each category. The area under the curve (AUC) ranged from 0.968 to 0.998.

Figure 3. Comparison of diagnostic performance between YOLO V.5 and ophthalmologists. (A) Confusion matrices of image numbers in YOLO V.5 to classify 36 anterior segment eye diseases into 9 categories using testing dataset of anterior segment photographs without clinical data. Confusion matrices of board-certified corneal specialist (B) and ophthalmology resident (C) without clinical information. (D) YOLO V.5 required 6.1 s to complete 500 image classifications, whereas 4118 s were required for corneal specialists and 3800 s were required for residents. IOL, intraocular lens.

Figure 4. Artificial intelligence (AI) performance and triage using smartphone images. (A) Comparison of AI performance (YOLO V.5) between slit-lamp (diffuser light) and smartphone images (336 patients). Owing to the various conditions of the smartphone (online supplemental figure S4A), the positive predictive values (PPV) were lower for the smartphone images (75.0%) than for the slit-lamp images (88.8%). (B) As the AI performance was better in images with a higher predictive score in the smartphone images, the AI performance in the images with a predictive score of 0.98 or greater was expected to exceed that of the slit-lamp images. (C) Confusion matrices of image numbers in YOLO V.5 for classifying 36 anterior segment eye diseases into 9 categories using smartphone images with a predictive score of 0.98 or greater. The PPV was 91.0%, which exceeded the 82.7% for board-certified corneal specialists. (D) Definition of triage classification. (E) Confusion matrices of image numbers in triage using smartphone images. High-performance triage was obtained after stratifying the images based on a predictive score greater than 0.98. IOL, intraocular lens.