A Deep Learning Model for Evaluating Meibomian Glands Morphology from Meibography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Meibomian gland dysfunction is a key factor in dry eye disease.
• Accurate segmentation of ocular structures is crucial for diagnosis and monitoring.
• Manual segmentation of meibography images is time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated segmentation of tarsus and meibomian gland areas.
• To assess the model's accuracy, sensitivity, and specificity in segmenting these ocular structures.
• To explore the potential clinical applications of automated segmentation in dry eye evaluation.

Main Methods:

• A U-net convolutional neural network architecture was employed.
• The model was trained on 1087 manually annotated meibography images from dry eye patients.
• Performance was evaluated using accuracy, sensitivity, specificity, and ROC curve analysis.

Main Results:

• The model achieved high accuracy for tarsus segmentation (0.985) with excellent sensitivity (0.975) and specificity (0.99).
• Meibomian gland segmentation demonstrated high accuracy (0.937) and specificity (0.96), with moderate sensitivity (0.751).
• Area under the curve (AUC) values were 0.985 for tarsus and 0.938 for meibomian glands.

Conclusions:

• The developed deep learning model accurately segments tarsus and meibomian gland areas from infrared meibography.
• The model shows significant potential for objective assessment of meibomian glands.
• Further refinement could enable widespread clinical application for dry eye assessment and research.