Development and multicenter validation of an AI driven model for quantitative meibomian gland evaluation.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Meibomian gland dysfunction (MGD) is a common condition affecting ocular surface health.
• Accurate quantitative evaluation of meibomian glands is crucial for diagnosing and managing MGD.
• Current methods for meibography image analysis can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) driven model for automated segmentation and quantitative evaluation of meibomian glands.
• To assess the performance of the AI model in comparison to conventional algorithms and manual grading.
• To evaluate the repeatability and external validity of the AI tool across multiple centers.

Main Methods:

• A multicenter retrospective study utilizing 1350 infrared meibography images from the Keratograph 5M device.
• Development and validation of an AI model for automated meibomian gland segmentation and quantification.
• Comparison of AI-based analysis with manual gland grading and counting, including assessment of repeatability and external validation.

Main Results:

• The AI model achieved high segmentation performance with an Intersection over Union of 81.67% and accuracy of 97.49%.
• Excellent agreement was observed between AI-based and manual gland grading (Kappa=0.93) and gland counting (Spearman r=0.9334).
• The model demonstrated stability, consistent external validation results (AUCs > 0.99), and outperformed conventional algorithms.

Conclusions:

• The developed AI tool provides a standardized, efficient, and objective method for analyzing meibography images.
• This AI-driven approach has the potential to enhance diagnostic precision for meibomian gland dysfunction.
• The tool may significantly assist in the clinical management of MGD across diverse patient populations.