Automatic Detection of Peripheral Retinal Lesions From Ultrawide-Field Fundus Images Using Deep Learning

Abstract

Purpose: To establish a multilabel-based deep learning (DL) algorithm for automatic detection and categorization of clinically significant peripheral retinal lesions using ultrawide-field fundus images. Methods: A total of 5958 ultrawide-field fundus images from 3740 patients were randomly split into a training set, validation set, and test set. A multilabel classifier was developed to detect rhegmatogenous retinal detachment, cystic retinal tuft, lattice degeneration, and retinal breaks. Referral decision was automatically generated based on the results of each disease class. t-distributed stochastic neighbor embedding heatmaps were used to visualize the features extracted by the neural networks. Gradient-weighted class activation mapping and guided backpropagation heatmaps were generated to investigate the image locations for decision-making by the DL models. The performance of the classifier(s) was evaluated by sensitivity, specificity, accuracy, F1 score, area under receiver operating characteristic curve (AUROC) with 95% CI, and area under the precision-recall curve. Results: In the test set, all categories achieved a sensitivity of 0.836–0.918, a specificity of 0.858–0.989, an accuracy of 0.854–0.977, an F1 score of 0.400–0.931, an AUROC of 0.9205–0.9882, and an area under the precision-recall curve of 0.6723–0.9745. The referral decisions achieved an AUROC of 0.9758 (95% CI= 0.9648–0.9869). The multilabel classifier had significantly better performance in cystic retinal tuft detection than the binary classifier (AUROC= 0.9781 vs 0.6112, P < 0.001). The model showed comparable performance with human experts. Conclusions: This new DL model of a multilabel classifier is capable of automatic, accurate, and early detection of clinically significant peripheral retinal lesions with various sample sizes. It can be applied in peripheral retinal screening in clinics.

Peripheral and posterior pole retinal changes in highly myopic Chinese children and adolescents

“…However, further research is required to explore the clinical implications and underlying mechanisms. For instance, deep learning algorithms may be used for the automatic detection and categorization of clinically significant retinal lesions [ 18 ].…”

Fundus Tessellated Density of Pathologic Myopia

“…Indeed, the advancements in artificial intelligence–assisted imaging processing have opened up new possibilities for semi-quantitative and quantitative analysis of fundus photographs and OCT scans for various ocular diseases, 16–19 and even possible to predict the progression and prognosis of myopia based on ocular biometrics 20 . However, there is a gap in the existing literature regarding the specific focus on FTD in patients with PM and the associated pathological changes observed in fundus photographs.…”

“…In addition, PM can lead to changes in the macular region, making it difficult to accurately determine whether patchy atrophy involves the macular area for patients in C3 and C4 based on color fundus photographs. Therefore, ex- Indeed, the advancements in artificial intelligence-assisted imaging processing have opened up new possibilities for semiquantitative and quantitative analysis of fundus photographs and OCT scans for various ocular diseases, [16][17][18][19] and even possible to predict the progression and prognosis of myopia based on ocular biometrics. 20 However, there is a gap in the existing literature regarding the specific focus on FTD in patients with PM and the associated pathological changes observed in fundus photographs.…”

Spotlight on Lattice Degeneration Imaging Techniques

“…In general, due to their notable morphological characteristics, including pigmentary changes and white lines, LD lesions have one of the highest AUROC following retinal detachment. 13 …”

“…In general, due to their notable morphological characteristics, including pigmentary changes and white lines, LD lesions have one of the highest AUROC following retinal detachment. 13 One more feasible deep learning system based on Optos Daytona images had AUROC of 0.888, 0.953 and 1.000 for detection of LD, retinal breaks, and retinal detachment, respectively. The referral accuracy was 79.8% compared to the reference standard.…”