Deep-OCTA: Ensemble Deep Learning Approaches for Diabetic Retinopathy Analysis on OCTA Images

The ultra-wide optical coherence tomography angiography (OCTA) has become an important imaging modality in diabetic retinopathy (DR) diagnosis. However, there are few researches focusing on automatic DR analysis using OCTA. In this paper, we present novel and practical deep-learning solutions based OCTA for the Diabetic Retinopathy Analysis Challenge (DRAC). first task of segmentation lesions, utilize UNet UNet++ to segment three lesions with strong data augmentation model ensemble. second image quality assessment, create ensemble Inception-V3, SE-ResNeXt, Vision Transformer models. Pre-training large dataset as well hybrid MixUp CutMix strategy both adopted boost generalization ability our third grading, build a find that pre-trained color fundus images serves useful substrate images. Extensive ablation studies demonstrate effectiveness each designed component solutions. proposed methods rank 4th, 3rd, 5th leaderboards DRAC, respectively. Our code is publicly available at https://github.com/FDU-VTS/DRAC .