Multicentre Pixel-Level Tear Meniscus Segmentation Dataset with Multimodal Imaging for Dry Eye Diagnosis

Abstract

Dry eye is one of the most common eye diseases and manifests as abnormalities in the quality, quantity, and fluid dynamics of tear fluid. Studying the secretion of tears is one of the methods for diagnosing dry eye. The lower tear meniscus height (TMH) is an important indicator of tear secretion and stability. This parameter is typically measured manually. Artificial intelligence (AI) can automatically segment and assess TMH images accurately. However, the success of AI models relies on high-quality datasets, including images and corresponding labels. Therefore, we introduced a multicentre, multimodal, pixel-level lower tear meniscus segmentation dataset. It comprised 1,693 colourful modal images and 1,739 infrared modal images from five centres across different regions of China, along with segmentation labels. These labels were generated using our newly developed human-computer interactive approach. To our knowledge, this is the only publicly available multimodal tear meniscus segmentation dataset. We believe this dataset will aid in constructing standardized medical image databases and advancing research on the diagnosis and treatment of dry eye.