FGADR

大规模的细粒度注释的糖尿病性视网膜病 (FGADR) 数据集由两个子集组成: Seg集 (1842图像) 和Grade集 (1000图像)。
糖尿病患者有患糖尿病视网膜病变 (DR) 的风险。这种疾病是当高血糖水平导致视网膜血管受损时。由于深度学习的巨大成功，计算机辅助DR诊断是早期检测DR和严重程度分级的有希望的选择。但是，由于缺乏具有一致且细粒度注释的训练数据，以前的大多数工作都提出了DR诊断系统，而对眼科医生而言却没有令人满意的性能或可解释性。为了解决这个问题，我们构造了一个包含2,842图像 (FGADR) 的大规模细粒度带注释的DR数据集。该数据集具有1,842幅图像，其中包含像素级DR相关病变注释，1,000幅图像级标签由6名板认证的眼科医生分级，具有评分者内部一致性。所提出的数据集能够对DR诊断进行广泛的研究。

The large-scale fine-grained annotated Diabetic Retinopathy (FGADR) dataset consists of two subsets: the Seg subset (1842 images) and the Grade subset (1000 images).

Patients with diabetes are at risk of developing diabetic retinopathy (DR), a condition caused by damage to retinal blood vessels due to high blood glucose levels. Thanks to the remarkable success of deep learning, computer-aided DR diagnosis has emerged as a promising approach for early DR detection and severity grading. However, most existing DR diagnosis systems fail to achieve satisfactory performance or interpretability for ophthalmologists, primarily due to the scarcity of training data with consistent and fine-grained annotations.

To address this issue, we constructed a large-scale fine-grained annotated DR dataset (FGADR) containing 2,842 images. Specifically, 1,842 of these images are equipped with pixel-level annotations of DR-related lesions, while the remaining 1,000 images have image-level labels graded by six board-certified ophthalmologists with good intra-rater consistency. The proposed dataset enables extensive research on DR diagnosis.

搜集汇总

背景与挑战

背景概述

FGADR是一个大规模细粒度注释的糖尿病性视网膜病变数据集，包含2842张图像，分为Seg集（1842张图像，提供像素级病变注释）和Grade集（1000张图像，提供由6名眼科医生分级的图像级标签）。该数据集旨在支持计算机辅助诊断研究，解决以往数据缺乏一致细粒度注释的问题，以提升糖尿病视网膜病变早期检测和严重程度分级的性能与可解释性。