Fundus-AVSeg：用于基于 AI 的动脉静脉分割的眼底图像数据集

抽象视网膜动脉静脉血管与全身性慢性疾病和心血管疾病有关。因此，视网膜动脉静脉血管的准确定量分析是临床诊断的初步依据。大多数现有的人工智能 （AI） 方法都是数据驱动的。虽然已经发布了一些公共视网膜动脉静脉分割数据集，但其数据质量并不令人满意。在本文中，我们为基于 AI 的动脉静脉分割建立了一个新的眼底图像数据集 Fundus-AVSeg。它由 100 张高分辨率眼底图像组成，并由专业眼科医生进行像素级手动注释。我们相信我们的 Fundus-AVSeg 将有利于视网膜动脉-静脉血管分割的进一步发展。<br>数据信息Fundus-AVSeg 包含 100 张眼底图像，其中 40 张来自正常眼底，20 张来自糖尿病视网膜病变 （DR） 患者，20 张来自年龄相关性黄斑变性 （AMD），20 张来自青光眼。图像有两种分辨率：2656×1992 和 1280×1280。像素级手动注释类别包括动脉、静脉、动脉-静脉交叉和类别不确定的血管。还为每张图像提供了两个图像质量类别：低质量和高质量。数据采集Fundus-AVSeg 的数据来源是深圳眼科医院。所有 100 张眼底图像均来自深圳市眼科医院的影像数据库。眼底图像由眼科医生使用蔡司VISUCAM200眼底相机或佳能眼底相机拍摄，这些都是眼底相机的主流产品。所有眼底图像都是在真实的临床诊断过程中生成的。所有伦理和实验程序和协议均由深圳眼科医院根据道德编号 2022KYPJ062 授予。数据记录Fundus-AVSeg 数据集已以压缩文件的形式上传到 Figshare。解压后的文件包含两个文件夹和一个Microsoft Office Excel 列表，以及两个 txt 格式的文件，分别命名为 ''images''、''annotation'' 、''metadata.xlsx''、'training.txt'' 和 ''testing.txt'' 。在 ''images'' 文件夹中，有 100 张眼底图像。图像被命名为 ''n\_D/A/G/N.png“，其中 ''n'' 表示眼底图像的数量，''D''、''A''、'G'' 和 ''N'' 代表 ''DR''、'AMD''、'青光眼'' 和 ''正常''。''annotation'' 文件夹包含 100 张对应的带注释的图像，这些图像按照相同的规则命名，其中该文件夹中的特定图像是 ''images'' 中同名图像的基本事实。“metadata.xlsx”是一个 Excel 文件，其中包含以下信息：图像名称、眼睛 ID、疾病类型和图像质量。''training.txt'' 和 ''testing.txt'' 文件分别存储用于训练和测试的特定图像名称，按照数据集的 8：2 分割。请注意，当前的数据拆分策略是我们推荐的，可以根据不同的研究目的进行更改。使用说明完整的数据集可通过提供的链接下载。用户可以根据其特定的研究设计灵活地划分数据集。预计用户将在他们的研究中引用这篇论文，并认识到数据集对他们研究的贡献。代码可用性本研究中提到的代码可以在 https://github.com/AI-thpremed/Basic-Seg-Experiment 中找到。<br><br>

Abstract: Retinal arteriovenous vessels are associated with systemic chronic diseases and cardiovascular diseases. Therefore, accurate quantitative analysis of retinal arteriovenous vessels serves as a preliminary basis for clinical diagnosis. Most existing artificial intelligence (AI) methods are data-driven. Although some public retinal arteriovenous segmentation datasets have been released, their data quality is unsatisfactory. In this paper, we establish a new fundus image dataset Fundus-AVSeg for AI-based arteriovenous segmentation. It consists of 100 high-resolution fundus images, with pixel-level manual annotations performed by professional ophthalmologists. We believe that our Fundus-AVSeg will facilitate further development of retinal arteriovenous vessel segmentation.

### Data Information
Fundus-AVSeg contains 100 fundus images, among which 40 are from normal fundi, 20 from patients with diabetic retinopathy (DR), 20 from age-related macular degeneration (AMD) patients, and 20 from glaucoma patients. The images have two resolutions: 2656×1992 and 1280×1280. The pixel-level manual annotation categories include arteries, veins, arteriovenous crossings, and vessels of uncertain category. Two image quality classes, low quality and high quality, are also provided for each image.

### Data Collection
The data of Fundus-AVSeg originates from Shenzhen Eye Hospital. All 100 fundus images are from the imaging database of Shenzhen Eye Hospital. The fundus images were captured by ophthalmologists using ZEISS VISUCAM 200 or Canon fundus cameras, which are mainstream fundus camera products. All fundus images were generated during real clinical diagnostic procedures. All ethical and experimental procedures and protocols were approved by Shenzhen Eye Hospital under ethical approval number 2022KYPJ062.

### Data Record
The Fundus-AVSeg dataset has been uploaded to Figshare as a compressed file. The decompressed file contains two folders, a Microsoft Office Excel spreadsheet, and two text files named "images", "annotation", "metadata.xlsx", "training.txt", and "testing.txt" respectively. In the "images" folder, there are 100 fundus images, named in the format "n_D/A/G/N.png", where "n" denotes the serial number of the fundus image, and "D", "A", "G", and "N" represent DR, AMD, glaucoma, and normal status respectively. The "annotation" folder contains 100 corresponding annotated images, named according to the same rule; the specific images in this folder serve as the ground truth for the identically named images in the "images" folder. The "metadata.xlsx" is an Excel file containing information such as image name, eye ID, disease type, and image quality. The "training.txt" and "testing.txt" files store the specific image names for training and testing respectively, following an 8:2 split of the dataset. Please note that the current data splitting strategy is our recommended one, and can be adjusted for different research purposes.

### Usage Instructions
The complete dataset can be downloaded via the provided link. Users can flexibly split the dataset according to their specific research designs. Users are expected to cite this paper in their research and acknowledge the contribution of this dataset to their work.

### Code Availability
The code mentioned in this study can be found at https://github.com/AI-thpremed/Basic-Seg-Experiment.

搜集汇总

背景与挑战

背景概述

Fundus-AVSeg是一个包含100张高分辨率眼底图像的数据集，专为基于AI的动脉静脉分割研究设计。这些图像来自正常眼底和三种眼部疾病患者，由专业眼科医生进行像素级手动注释，并提供了详细的元数据和推荐的数据划分策略。数据集以CC BY 4.0许可发布，支持灵活的研究用途。