Image based analysis of meibomian gland dysfunction using conditional generative adversarial neural network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Meibomian gland dysfunction (MGD) is a leading cause of dry eye disease.
• Current diagnostic methods struggle to quantify MGD irregularities in infrared (IR) images, including issues with light reflection, boundaries, focus, and positioning.
• Accurate analysis of MGD severity and variations is crucial in clinical settings.

Purpose of the Study:

• To develop and validate a novel model for precise quantification and analysis of MGD from IR images.
• To overcome limitations of existing methods in analyzing MG irregularities and segmentation.
• To improve the detection and analysis of MGD dropout areas.

Main Methods:

• A conditional generative adversarial network (GAN) model was proposed, utilizing adversarial learning.
• The generator maps IR images to a confidence map, while the discriminator penalizes mismatches.
• Confidence maps were thresholded to segment Meibomian glands (MGs) and intergland boundaries from IR images.

Main Results:

• The model demonstrated significant improvement in quantifying IR image irregularities.
• Performance was evaluated using meiboscoring, grading, Pearson correlation, and Bland-Altman analysis.
• Quality assessment included average Pompeiu-Hausdorff distance and Aggregated Jaccard Index.

Conclusions:

• The developed technique offers enhanced quantification of IR image irregularities for MGD analysis.
• This method surpasses state-of-the-art results in detecting and analyzing MGD dropout areas.
• The approach provides a more accurate and reliable tool for MGD assessment in clinical practice.