Introduction: The Importance of Meibomian Gland Health in Ocular Surface Disease

The meibomian glands are integral to ocular surface homeostasis, secreting the lipids that stabilize the tear film and reduce tear evaporation. Their dysfunction—known as meibomian gland dysfunction (MGD)—is recognized as the leading cause of evaporative dry eye disease and contributes significantly to ocular surface disorders, including post-surgical complications. MGD manifests as altered gland morphology, reduced lipid secretion, and gland dropout, leading to symptomatic dry eye and chronic ocular discomfort.

Traditional assessments of meibomian gland health have relied heavily on subjective evaluations using slit-lamp biomicroscopy and manual grading of meibography images. These approaches are limited by poor reproducibility, significant interobserver variability, and labor-intensive workflows, particularly in high-volume or resource-constrained clinical settings.

Recent advances in artificial intelligence (AI), particularly deep learning techniques such as convolutional neural networks (CNNs), U-Net, and EfficientNet architectures, offer a solution to these limitations. These models are capable of automated, objective, and reproducible analysis of meibography images, enabling precise quantification of gland morphology, dropout, and structural changes. By enhancing diagnostic accuracy and facilitating longitudinal monitoring, AI-driven image analysis represents a significant shift in the evaluation and management of ocular surface diseases. Growing evidence from multicenter studies confirms the reliability and scalability of these tools, setting the stage for widespread clinical integration.

Multicenter Validation: Testing Generalizability

To assess the robustness and generalizability of AI-driven meibomian gland evaluation, a multicenter validation study was conducted across three distinct clinical sites in two countries, employing multiple imaging systems including the Keratograph and LipiView platforms. The datasets encompassed a broad spectrum of patient demographics—spanning age, ethnicity, and disease severity—and included variations in image quality, reflecting real-world clinical heterogeneity.

All images were annotated by expert clinicians to establish a consistent reference standard. The AI model was then evaluated against these human annotations, with segmentation performance quantified using interclass correlation coefficients (ICC) and kappa statistics. The results demonstrated high concordance with manual grading, confirming the model’s accuracy and reliability across imaging platforms and patient populations.

Importantly, the model processed each image in under a second, achieving substantial time savings and workflow efficiencies compared to manual assessment. This speed and consistency render the system highly suitable for high-throughput environments and enhance its utility in telemedicine and remote diagnostic contexts. The study’s findings underscore the model’s capacity to standardize meibomian gland assessment at scale, bridging gaps in expertise and access across varied clinical settings.

Clinical Applications: From Research to Routine Care

The clinical utility of AI-driven meibomian gland analysis extends far beyond academic research, offering tangible benefits in everyday practice. By quantitatively measuring key gland parameters—such as area, length, density, and dropout patterns—the model enables early detection of glandular alterations and supports objective monitoring of disease progression and therapeutic response.

This precision is particularly valuable in diagnosing and managing MGD, where conventional subjective methods often fail to detect subtle morphological changes or to consistently track treatment outcomes. Integration of AI-generated metrics with electronic health records enhances documentation and longitudinal follow-up, while compatibility with mobile diagnostic devices and telemedicine platforms broadens access in underserved regions.

Furthermore, the model’s standardized output facilitates data harmonization in multicenter clinical trials, reducing variability and improving reproducibility across different populations and imaging systems. Validation results showing accuracy rates above 97% and strong correlations with expert assessment indicate that the technology is ready for routine deployment in clinical settings.

By automating the evaluation of meibography, the AI framework minimizes observer bias, streamlines workflow, and enhances diagnostic precision. These capabilities collectively support a more personalized and data-driven approach to ocular surface disease management, positioning AI as a central tool in the evolution of modern ophthalmology.

Reference:

  1. Chhadva P, Goldhardt R, Galor A. Meibomian Gland Disease: The Role of Gland Dysfunction in Dry Eye Disease. Ophthalmology. 2017 Nov;124(11S):S20-S26. doi: 10.1016/j.ophtha.2017.05.031. PMID: 29055358; PMCID: PMC5685175.

  2. Kaur K, Stokkermans TJ. Meibomian Gland Disease. [Updated 2024 Mar 3]. In: StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing; 2025 Jan-. Available from: https://www.ncbi.nlm.nih.gov/books/NBK580474/

  3. SULLIVAN, D., KAM, W., KHANDELWAL, P. and LIU, S. (2010), Control of the Meibomian gland in health and disease. Acta Ophthalmologica, 88: 0-0. https://doi.org/10.1111/j.1755-3768.2010.2435.x

  4. Li L, Xiao K, Shang X, Hu W, Yusufu M, Chen R, Wang Y, Liu J, Lai T, Guo L, Zou J, van Wijngaarden P, Ge Z, He M, Zhu Z. Advances in artificial intelligence for meibomian gland evaluation: A comprehensive review. Surv Ophthalmol. 2024 Nov-Dec;69(6):945-956. doi: 10.1016/j.survophthal.2024.07.005. Epub 2024 Jul 23. PMID: 39025239.

  5. Zhang Y-Y, Zhao H, Lin J-Y, Wu S-N, Liu X-W, Zhang H-D, Shao Y and Yang W-F (2021) Artificial Intelligence to Detect Meibomian Gland Dysfunction From in-vivo Laser Confocal Microscopy. Front. Med. 8:774344. doi: 10.3389/fmed.2021.774344

  6. Zhang, Z.; Lin, X.; Yu, X.; Fu, Y.; Chen, X.; Yang, W.; Dai, Q. Meibomian Gland Density: An Effective Evaluation Index of Meibomian Gland Dysfunction Based on Deep Learning and Transfer Learning. J. Clin. Med. 2022, 11, 2396. https://doi.org/10.3390/jcm11092396

  7. li li, Kunhong Xiao, Hao Huang, Lingyi Liang, Ying Xue, Zhuoting Zhu; AI-Driven Model for Quantitative Evaluation of Meibomian Glands: Novel Development and Multicenter Validation. Invest. Ophthalmol. Vis. Sci. 2025;66(8):3852.

  8. Graham, A.D., Wang, J., Kothapalli, T. et al. Artificial intelligence models utilize lifestyle factors to predict dry eye related outcomes. Sci Rep 15, 13378 (2025). https://doi.org/10.1038/s41598-025-96778-x

  9. Forni R, Maruotto I, Zanuccoli A, Nicoletti R, Trimigno L, Corbellino M, Travé-Huarte S, Giannaccare G, Gargiulo P. Advancing Meibography Assessment and Automated Meibomian Gland Detection Using Gray Value Profiles. Diagnostics. 2025; 15(10):1199. https://doi.org/10.3390/diagnostics15101199

  10. Nilufer Yesilirmak, Volkan Okbay, Yener Yesilirmak, Omer Mustafa Bilgic, Oyku Diribas, Onat Akca, Gozde B. Akar, Evaluation of meibomian gland dysfunction with deep learning model considering different datasets and gland morphology, Computers in Biology and Medicine,