A Deep Learning Algorithm for Classifying Diabetic Retinopathy Using Optical Coherence Tomography Angiography

Abstract

Purpose To develop an automated diabetic retinopathy (DR) staging system using optical coherence tomography angiography (OCTA) images with a convolutional neural network (CNN) and to verify the feasibility of the system. Methods In this retrospective cross-sectional study, a total of 918 data sets of 3 × 3 mm 2 OCTA images and 917 data sets of 6 × 6 mm 2 OCTA images were obtained from 1118 eyes. A deep CNN and four traditional machine learning models were trained with annotations made by a retinal specialist based on ultra-widefield fluorescein angiography. Separately, the same images of the test data sets were independently graded by two human experts. The results of the CNN algorithm were compared with those of traditional machine learning–based classifiers and human experts. Results The proposed CNN achieved an accuracy of 0.728, a sensitivity of 0.675, a specificity of 0.944, an F1 score of 0.683, and a quadratic weighted κ of 0.908 for a six-level staging task, which were far superior to the results of traditional machine learning methods or human experts. The CNN algorithm showed a better performance using 6 × 6 mm 2 rather than 3 × 3 mm 2 sized OCTA images and using combined data rather than a separate OCTA layer alone. Conclusions CNN-based classification using OCTA images can provide reliable assistance to clinicians for DR classification. Translational Relevance This CNN algorithm can guide the clinical decision for invasive angiography or referrals to ophthalmology specialists, helping to create more efficient diagnostic workflow in primary care settings.

Artificial intelligence and hemodynamic studies in optical coherence tomography angiography for diabetic retinopathy evaluation: A review

“…73,74 Certain investigations of DL and ML frameworks in DR screening using OCTA Various ML and DL techniques were implemented and available to diagnose and classify DR automatically using fundus image but few limitations are there in those modeled algorithms as a fundus image is a 2D image where deep blood vessel features are missing, it produces an image with poor resolution so, hemodynamic analysis using ML technique is not possible. [75][76][77] Various DL frameworks and ML algorithms using OCT angiography images for screening and classifying DR were considered for this review.…”

“…Among various models, the system preferred the ResNet 101 CNN model classifier in stage 1 and achieved more accuracy with 91%-98%, 86%-97% sensitivity, and 94%-99% specificity compared to the other two results in performance analysis. 77…”

“…The performance indicators in terms of accuracy, sensitivity and specificity of various ML classification models are plotted in bar chart. 58,77 Performance indicators for human expert prediction of DR classification show 33% accuracy, 32% sensitivity, and 86.9% specificity. The results are considered to be very low compared to ML techniques considered for the analysis in this article as shown in Figure 8.…”

DeepVasNet: Retinal Vasculature and FAZ Segmentation in OCTA with EfficientNet-Enhanced U-Net

“…Various methods are available in the state-of-the-art literature for retinal blood vessels and FAZ segmentation. This includes delineation of the retinal vasculature alone [1,4,11,16] or combined retinal vessel and FAZ detection [5,14,15,[17][18][19]. Some studies further segment the retinal blood vessels into arteries, veins, and capillaries [2,8,20].…”

“…This method was validated on two public OCTA databases, OCTAGON and sFAZDATA, to record comparable performance with other deep learning models. Further, Ryu et al [14] implemented multi-stage segmentation of retinal images. The first phase of the work dealt with retinal blood vessels and FAZ segmentation.…”