Fairer AI in Ophthalmology via Implicit Fairness Learning for Mitigating Sexism and Ageism

We collect the largest and most diverse fundus image dataset with data from over 8,405 patients representing a wide age range (0 to 90 years). The fundus dataset contains two types of advanced ultra-widefield and regular narrow-angle fundus images, with the ultra-widefield imaging dataset containing 16,530 fundus images annotated with 38 ophthalmic diseases and 67 fundus features and the narrow-angle imaging dataset containing 4,540 fundus images annotated with 16 ophthalmic diseases and 20 fundus features. We hereby only provide the test set of our dataset, while the complete dataset can be accessed through the Google Drive link provided in the paper. (Despite our attempts to upload the full dataset, the transmission time was excessively long due to its size exceeding 100G, necessitating this alternative method.)

本研究采集了目前规模最大、多样性最丰富的眼底图像（fundus image）数据集，数据来自8405余名患者，年龄覆盖0至90岁的宽泛区间。该数据集包含两类成像类型：先进超宽域（ultra-widefield）眼底图像与常规窄角眼底图像。其中超宽域成像子集包含16530张眼底图像，标注有38种眼科疾病与67种眼底特征；窄角成像子集则包含4540张眼底图像，标注有16种眼科疾病与20种眼底特征。本研究仅公开该数据集的测试集，完整数据集可通过论文中提供的Google Drive链接获取。（由于完整数据集体积超过100GB，上传传输耗时过长，尽管我们尝试过直接上传完整数据集，最终只能采用此替代方案。）