An ultra-wide-field fundus image dataset for intelligent diagnosis of intraocular tumors

Retinal fundus photographs are now widely used in developing artificial intelligence (AI) systems for the detection of various fundus diseases. However, the application of AI algorithms in intraocular tumors remains limited due to the scarcity of large, publicly available, and diverse multi-disease fundus image datasets. Ultra-wide-field (UWF) fundus images, with a broad imaging range, offer unique advantages for early detection of intraocular tumors. In this study, we developed a comprehensive dataset containing 2,031 UWF fundus images, encompassing five distinct types of intraocular tumors and normal images. These images were classified by three expert annotators to ensure accurate labeling. This dataset aims to provide a robust, multi-disease, real-world data source to facilitate the development and validation of AI models for the early diagnosis and classification of intraocular tumors.