Fundus-AVSeg

AbstractRetinal artery-vein vessels are associated with systemic chronic diseases and cardiovascular diseases. Therefore, the accurate quantitative analysis of retinal artery-vein vessels is the preliminary basis of clinical diagnosis. Most of the existing artificial intelligence(AI) methods are data-driven. Although some public retinal artery-vein vessel segmentation datasets have been released, their data quality is unsatisfactory. In this paper, we establish a new fundus image dataset for AI-based artery-vein segmentation, Fundus-AVSeg. It consists of 100 high-resolution fundus images with pixel-wise manual annotation by professional ophthalmologists. We believe our Fundus-AVSeg will benefit the further development of retinal artery-vein vessel segmentation.<br>Data Information Fundus-AVSeg contains 100 fundus images, 40 of which are from normal fundus, 20 from diabetic retinopathy(DR) patients, 20 from age-related macular degeneration (AMD), and 20 from glaucoma. Images have two resolutions: 2656×1992 and 1280×1280. Pixel-wise manual annotated categories include arteries, veins, artery-vein crossings, and vessels of uncertain category. Two image quality categories are also provided for each image: low-quality and high-quality. Data CollectionThe data source of Fundus-AVSeg is Shenzhen Eye Hospital. All 100 fundus images are obtained from the imaging database of Shenzhen Eye Hospital. The fundus images are captured by ophthalmologists using ZEISS VISUCAM200 fundus cameras or Canon fundus cameras, which are the mainstream products of fundus cameras. All fundus images are generated during real clinical diagnostic processes. Approval of all ethical and experimental procedures and protocols is granted by the Shenzhen Eye Hospital under the ETHICAL NUMBER 2022KYPJ062. Data RecordsThe Fundus-AVSeg dataset has been uploaded to Figshare in the form of a zipped file. The unzipped file contains two folders and one Microsoft Office Excel list, and two txt format files, named ``images'', ``annotation'' , ``metadata.xlsx'', ``training.txt'', and ``testing.txt'', respectively. In the ``images'' folder, there are 100 fundus images. Images are named ``n\_D/A/G/N.png", where ``n'' means the number of fundus images and ``D'', ``A'', ``G'', and ``N'' stand for ``DR'', ``AMD'', ``Glaucoma'', and ``Normal''. The ``annotation'' folder contains 100 corresponding annotated images, which are named according to the same rule, where a specific image in this folder is the ground truth of the image with the same name in the ``images''. The ``metadata.xlsx'' is an Excel file that holds the following information: 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 split strategy is recommended by us and can be altered for different research purposes.Usage NotesThe complete dataset is available for download via the provided link. Users have the flexibility to divide the dataset based on their specific study designs. It is expected that users will reference this paper in their research and recognize the dataset’s contribution to their studies.Code AvailabilityThe code mentioned in this study can be found at https://github.com/AI-thpremed/Basic-Seg-Experiment.<br><br>