[An advanced imaging method for measuring and assessing meibomian glands based on deep learning]

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Meibomian gland dysfunction (MGD) is a common cause of dry eye disease.
• Accurate and efficient evaluation of meibomian gland loss is crucial for diagnosis and management.
• Traditional methods for assessing meibomian glands can be time-consuming and subjective.

Purpose of the Study:

• To assess the clinical utility of a deep learning-based imaging technique for rapid meibomian gland measurement.
• To evaluate the accuracy of a deep learning model in identifying and quantifying meibomian gland loss.
• To compare the efficiency of the deep learning method with manual evaluation by clinicians.

Main Methods:

• A diagnostic evaluation study collected 2,304 meibomian gland images from 576 dry eye patients.
• A deep learning algorithm was developed and trained on labeled images to identify meibomian glands and calculate loss.
• Model performance was assessed using mean average precision (mAP) and validation loss; results were compared to independent clinician evaluations.

Main Results:

• The deep learning model demonstrated high accuracy in marking meibomian conjunctiva (mAP>0.976) and meibomian glands (mAP>0.922).
• Calculated meibomian gland loss proportions by the model (53.24%±11.09%) showed no significant difference compared to manual marking (52.13%±13.38%).
• The model evaluated each image in under 0.5 seconds, drastically faster than the >10 seconds required by clinicians.

Conclusions:

• The deep learning-based imaging model significantly improves the accuracy and efficiency of meibomian gland examination.
• This AI-driven approach offers a valuable tool for auxiliary diagnosis and screening of MGD-related diseases.
• The method's speed and precision support its application in busy clinical settings for dry eye management.