AI driven quantitative analysis of meibomian glands in children and adolescents: a benchmark dataset study

Due to the lack of quality-controlled quantitative data on meibomian gland (MG) morphology in children and adolescents, this study aims to establish a Children and Adolescents Meibomian Gland (CAMG) dataset.

A total of 1114 quality-controlled upper eyelid infrared images were collected from 730 children and adolescent subjects using the Oculus Keratograph 5 M. Images underwent preprocessing and multi-stage expert quality control screening before segmentation. Morphological parameters including gland area, gland dropout ratio, gland length and width, number of glands, and total glands ratio were extracted using an AI model. The dataset, comprising images, annotations, and demographic information, is openly accessible on Figshare, with AI model codes available on GitHub to support research reproducibility and algorithm optimization.

The dataset includes 1114 high-resolution quality-controlled upper eyelid images from 730 subjects (mean age 11.80 ± 2.39 years; 46.77% male), accompanied by AI-assisted segmentation annotations and corresponding morphological measurements. The U-Net segmentation model achieved an accuracy of 97.49%, a Dice coefficient of 89.72%, and an intersection over union (IoU) of 81.67%. Quantitative analysis revealed that MG parameters remained relatively stable in adolescents compared to children. Females exhibited significantly wider and larger MGs than males. Similar sex-related differences were also observed in the central five MGs. Males exhibited a higher MG count compared to females.

CAMG is a publicly available MG dataset for children and adolescents to support AI-based individualized clinical assessments. The dataset's transparent quality control processes establish a foundation for epidemiological research, promoting cross-institutional collaboration and AI-driven advancements in ophthalmology.