An Ensemble Method to Automatically Grade Diabetic Retinopathy with Optical Coherence Tomography Angiography Images

Diabetic retinopathy (DR) is a complication of diabetes, and one the major causes vision impairment in global population. As early-stage manifestation DR usually very mild hard to detect, an accurate diagnosis via eye-screening clinically important prevent loss at later stages. In this work, we propose ensemble method automatically grade using ultra-wide optical coherence tomography angiography (UW-OCTA) images available from Retinopathy Analysis Challenge (DRAC) 2022. First, adopt state-of-the-art classification networks, i.e., ResNet, DenseNet, EfficientNet, VGG, train them UW-OCTA with different splits dataset. Ultimately, obtain 25 models, which, top 16 models are selected ensembled generate final predictions. During training process, also investigate multi-task learning strategy, add auxiliary task, Image Quality Assessment, improve model performance. Our achieved quadratic weighted kappa (QWK) 0.9346 Area Under Curve (AUC) 0.9766 on internal testing dataset, QWK 0.839 AUC 0.8978 DRAC challenge