Purpose

The purpose of this study was to develop an unsupervised feature learning approach that automatically measures Meibomian gland (MG) atrophy severity from meibography images and discovers subtle relationships between meibography images according to visual similarity.

Methods

One of the latest unsupervised learning approaches is to apply feature learning based on nonparametric instance discrimination (NPID), a convolutional neural network (CNN) backbone model trained to encode meibography images into 128-dimensional feature vectors. The network aims to learn a similarity metric across all instances (e.g. meibography images) and groups visually similar instances together. A total of 706 meibography images with corresponding meiboscores were collected and annotated for the use of network learning and performance evaluation.

Results

Four hundred ninety-seven meibography images were used for network learning and tuning, whereas the remaining 209 images were used for network model evaluations. The proposed nonparametric instance discrimination approach achieved 80.9% meiboscore grading accuracy on average, outperforming the clinical team by 25.9%. Additionally, a 3D feature visualization and agglomerative hierarchical clustering algorithms were used to discover the relationship between meibography images.

Conclusions

The proposed NPID approach automatically analyses MG atrophy severity from meibography images without prior image annotations, and categorizes the gland characteristics through hierarchical clustering. This method provides quantitative information on the MG atrophy severity based on the analysis of phenotypes.

Translational relevance

The study presents a Meibomian gland atrophy evaluation method for meibography images based on unsupervised learning. This method may be used to aid diagnosis and management of Meibomian gland dysfunction without prior image annotations, which require time and resources.

PURPOSE: This study enhances Meibomian gland (MG) infrared image analysis in dry eye (DE) research through artificial intelligence (AI). It is comprised of two main stages: automated eyelid detection and tarsal plate segmentation to standardize meibography image analysis. The goal is to address limitations of existing assessment methods, bridge the curated and real-world dataset gap, and standardize MG image analysis. METHODS: The approach involves a two-stage process: automated eyelid detection and tarsal plate segmentation. In the first stage, an AI model trained on curated data identifies relevant eyelid areas in non-curated datasets. The second stage refines the eyelid area in meibography images, enabling precise comparisons between normal and DE subjects. This approach also includes specular reflection removal and tarsal plate mask refinement. RESULTS: The methodology achieved a promising instance-wise accuracy of 80.8% for distinguishing meibography images from 399 DE and 235 non-DE subjects. By integrating diverse datasets and refining the area of interest, this approach enhances meibography feature extraction accuracy. Dimension reduction through Uniform Manifold Approximation and Projection (UMAP) allows feature visualization, revealing distinct clusters for DE and non-DE phenotypes. CONCLUSIONS: The AI-driven methodology presented here quantifies and classifies meibography image features and standardizes the analysis process. By bootstrapping the model from curated datasets, this methodology addresses real-world dataset challenges to enhance the accuracy of meibography image feature extraction. TRANSLATIONAL RELEVANCE: The study presents a standardized method for meibography image analysis. This method could serve as a valuable tool in facilitating more targeted investigations into MG characteristics.