Automated multidimensional deep learning platform for referable diabetic retinopathy detection: a multicentre, retrospective study

Abstract

ObjectiveTo develop and validate a real-world screening, guideline-based deep learning (DL) system for referable diabetic retinopathy (DR) detection.DesignThis is a multicentre platform development study based on retrospective, cross-sectional data sets. Images were labelled by two-level certificated graders as the ground truth. According to the UK DR screening guideline, a DL model based on colour retinal images with five-dimensional classifiers, namely image quality, retinopathy, maculopathy gradability, maculopathy and photocoagulation, was developed. Referable decisions were generated by integrating the output of all classifiers and reported at the image, eye and patient level. The performance of the DL was compared with DR experts.SettingDR screening programmes from three hospitals and the Lifeline Express Diabetic Retinopathy Screening Program in China.Participants83 465 images of 39 836 eyes from 21 716 patients were annotated, of which 53 211 images were used as the development set and 30 254 images were used as the external validation set, split based on centre and period.Main outcomesAccuracy, F1 score, sensitivity, specificity, area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), Cohen’s unweighted κ and Gwet’s AC1 were calculated to evaluate the performance of the DL algorithm.ResultsIn the external validation set, the five classifiers achieved an accuracy of 0.915–0.980, F1 score of 0.682–0.966, sensitivity of 0.917–0.978, specificity of 0.907–0.981, AUROC of 0.9639–0.9944 and AUPRC of 0.7504–0.9949. Referable DR at three levels was detected with an accuracy of 0.918–0.967, F1 score of 0.822–0.918, sensitivity of 0.970–0.971, specificity of 0.905–0.967, AUROC of 0.9848–0.9931 and AUPRC of 0.9527–0.9760. With reference to the ground truth, the DL system showed comparable performance (Cohen’s κ: 0.86–0.93; Gwet’s AC1: 0.89–0.94) with three DR experts (Cohen’s κ: 0.89–0.96; Gwet’s AC1: 0.91–0.97) in detecting referable lesions.ConclusionsThe automatic DL system for detection of referable DR based on the UK guideline could achieve high accuracy in multidimensional classifications. It is suitable for large-scale, real-world DR screening.

Is it time to consider teleophthalmology as a game-changer in the management of diabetic retinopathy?

“…The authors emphasized a need for meticulous testing of all algorithms before clinical implementation. The latest validation retrospective study conducted by Zhang et al (37) underscored that developed automatic DL system for referable DR detection is suitable for real-word screening.…”

Cost-effectiveness analysis of telemedicine and artificial intelligence-based diabetic retinopathy screening in urban and rural China

“…[26,27] The grading accuracy of manual grading in DR diagnosis was based on a recent pooled analysis, and the AI-based grading accuracy was based on our previous studies. [11,15,28] The compliance of opportunistic case nding, screening, and treatment in rural and urban were derived from a previous report. [18] All parameters were shown in Table 1.…”

“…Telemedicine can assist the DR screening by referring those individuals classi ed as DR to the tertiary centers for the detail ophthalmic examinations and timely treatment. [10] Recently, arti cial intelligence (AI) with the deep learning (DL) algorithm showed high accuracy in the identi cation of DR [11] . AI should be able to assist the DR screening [12][13][14] as AI can reduce the availability of human assessors and provide long-term sustainability for DR screening.…”

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

“…18,19 Deep learning (DL) has been applied in the detection and screening of different ocular diseases, including diabetic retinopathy, glaucoma, age-related macular degeneration, and other retinal diseases. [20][21][22][23][24][25][26][27] Convolutional neural networks (CNNs) as a DL algorithm are suitable for image recognition 28,29 and have been applied for RRD identification based on the UWF fundus images. Ohsugi et al 30 implemented a CNN-based model on 831 UWF images for RRD detection with a sensitivity of 97.6% and a specificity of 96.5%.…”