A multi-category eye image dataset for AI-based segmentation and analysis of eyelid disorders

Abnormal eyelid position and morphology can cause visual dysfunction, ocular surface damage, and facial deformities, highlighting the need for accurate eyelid assessment. Advances in artificial intelligence (AI) have enabled automated analysis of eyelid abnormalities using external eye images, but development is limited by the lack of publicly available datasets with multi-category disorders and standardized structural annotations. To address this gap, we constructed a clinically annotated eye image dataset of 1,414 images from eight common eyelid disorders and a normal control group. Each image includes diagnostic labels and manual annotations of three key periocular structures: eyelid fissure, cornea, and eyebrow. Image quality evaluation and expert verification were performed to ensure annotation reliability. Inter- and intra-annotator consistency demonstrated excellent agreement. A baseline Attention 2D U-Net segmentation model trained on the dataset achieved Dice coefficients of 0.93, 0.96, and 0.89 for the eyelid fissure, cornea, and eyebrow, respectively. This dataset provides a valuable resource for automated segmentation, quantitative periocular measurement, and AI-assisted eyelid analysis, supporting standardized and reproducible approaches for eyelid assessment.