Graham Andrew D, Yeh Chun-Hsiao, Lin Meng C
Clinical Research Center, University of California, Berkeley, Berkeley, CA, USA.
Vision Science Group, Herbert Wertheim School of Optometry and Vision Science, University of California, Berkeley, Berkeley, CA, USA.
Transl Vis Sci Technol. 2025 Aug 1;14(8):36. doi: 10.1167/tvst.14.8.36.
To determine the impact of age, sex, ethnicity, and contact lens wear on the detailed morphology of the Meibomian glands as quantified by a deep learning segmentation model.
A large dataset of meibography images (n = 2233) from 560 subjects was compiled and input to a supervised machine learning model to quantify gland length, width, tortuosity, contrast, atrophy, density, and number of glands. These morphology outcomes were modeled as functions of age, sex, ethnicity, and contact lens wear parameters.
Age was significantly associated with shorter glands, more atrophy, and lower gland density (all p < 0.001). No Meibomian gland morphological characteristics were related to sex. Asian subjects exhibited the longest and densest glands, and black subjects exhibited the most gland atrophy. Although contact lens wearers overall had significantly longer glands (∼4%-5%; p < 0.001) than non-wearers, no other contact lens wear parameter was significantly related to any Meibomian gland morphological feature.
What constitutes "normal-looking" Meibomian glands in a meibography image depends on the age and ethnicity of the patient. There appear to be no significant female/male Meibomian gland morphological differences. Meibomian gland morphology is robust to contact lens wear based on a large-sample analysis of younger, successful contact lens wearers. Now that the impact of these external factors has been established, work is ongoing to determine exactly what alterations in Meibomian gland morphology contribute to downstream pathology.
This study used artificial intelligence to provide clinicians with novel insights into normal versus abnormal Meibomian gland morphological features in their patients.
通过深度学习分割模型量化睑板腺的详细形态，以确定年龄、性别、种族和隐形眼镜佩戴情况对其的影响。
收集了560名受试者的大量睑板腺图像数据集（n = 2233），并将其输入到一个监督机器学习模型中，以量化腺体长度、宽度、曲折度、对比度、萎缩程度、密度和腺体数量。这些形态学结果被建模为年龄、性别、种族和隐形眼镜佩戴参数的函数。
年龄与较短的腺体、更多的萎缩和较低的腺体密度显著相关（所有p < 0.001）。睑板腺的形态特征与性别无关。亚洲受试者的腺体最长且最密集，而黑人受试者的腺体萎缩最为明显。虽然总体上隐形眼镜佩戴者的腺体比非佩戴者明显更长（约4%-5%；p < 0.001），但没有其他隐形眼镜佩戴参数与任何睑板腺形态特征显著相关。
睑板腺图像中“外观正常”的睑板腺取决于患者的年龄和种族。男女睑板腺形态似乎没有显著差异。基于对年轻且成功佩戴隐形眼镜者的大样本分析，睑板腺形态对隐形眼镜佩戴具有较强的耐受性。既然已经确定了这些外部因素的影响，目前正在开展工作以确定睑板腺形态的哪些具体改变会导致下游病理变化。
本研究使用人工智能为临床医生提供关于患者睑板腺正常与异常形态特征的新见解。
