A study recently published in Translational Vision Science & Technology evaluated the performance of a deep learning-based interactive application in analyzing meibography images to understand the progression of meibomian gland dysfunction (MGD).
Give me some background.
MGD is present in over 85% of patients with dry eye disease and currently affects between 16 and 49 million people in the United States, or approximately 5-15% of the population.
Meibography is the primary imaging modality used to assess meibomian glands (MGs), and the key feature analyzed is gland loss area, as shortening and atrophy of MGs are an observable pathology of MGD.
I’m sensing a but …
However: Manually determining the area of gland loss relative to the tarsal plate is subjective and can leave room for intra- and integrated variability.
Software, such as ImageJ (National Institutes of Health) can outline the loss area to measure percentage, but this process is time consuming and varies based on the quality and focus of the meibography images.
What about this new deep learning-based application?
Despite recent advancements in automated and objective methods to determine MG loss and morphology, existing methods feature notable limitations, including:
- Largely focusing only on upper eyelids
- Lacking comprehensive editing features after automatic inference
- Relying on ideal high-quality images with refined regions of interest
Gotcha. And so …
To address these gaps, researchers developed an interactive image editor—Ophthalmic Segmentation and Analysis Software (OASIS)—that features:
- Deep learning-based eyelid and gland segmentation models
- Manual editing tools
- Quantitative metric generation
How it works: Three anatomically derived masks (eyelid, gland, and gland loss) are annotated manually or with model assistance, followed by quantitative metric generation to evaluate MG area.
Thoma Stokkermans, OD, PhD, one of the study authors, noted that OASIS “will enhance our ability to do research on dry eye, especially to help us understand the natural history and progression of MG loss and dysfunction [as well as] study how dry eye interventions affect the MGs.”
Now talk about the study.
In this natural history study, the research team collected 2,439 meibography images from 325 patients across 11 sites using the LipiView II device (Johnson & Johnson).
- Each participant underwent imaging at an initial visit and a follow-up 90 days later.
Clinicians used OASIS for image analysis, which involved both manual and deep-learning assisted processes. In the manual process, clinicians annotated three distinct masks per image, including:
- The eyelid
- Glands
- Gland loss
And in the assisted process?
OASIS incorporated deep-learning models to infer gland masks, reducing the time required for gland-by-gland annotation.
The software’s interface provided additional tools for image enhancement and calculation of currently used clinical metrics—including the Pult scale.
Now to the findings.
OASIS enabled clinicians to quantitatively analyze MGD in under 3 minutes versus 15-20 minutes with traditional manual analysis, representing an 87% reduction in time.
The software accurately calculated Pult meiboscale grades with fair agreement between the clinician and software (kappa = 0.79)—demonstrating a high level of consistency.
- Note: The kappa is one of the most commonly used statistics to test interrater reliability and can range from -1 to +1.
Expert opinion?
“The emphasis of all meibography interpretations is on MG loss, noted using the five-point Pult scale, or a percentage loss, but that there are many other morphological changes that develop in MGD,” added Dr. Stokkermans.
And how could this be used in the future?
OASIS may be used to analyze MG morphometrics, including the 12 additional meibographical signs of MGD besides gland shortening identified in the 2020 DREAM study, such as:
- Gland tortuosity
- Distortion
- Hooked glands
- Drop-out
- Thickening
- Thinning
- Overlapping
- Ghosting
- Tadpoling
- Abnormal gap formation
- Fluffy areas
- No extension to the lid margin
Any limitations?
These included:
- This study only evaluated the performance of the gland model and focused solely on binary segmentation
- Future investigations could benefit from evaluating multiple models for both glands and eyelids with varied parameters and architectures
- Lid eversion and blurring were common issues in the annotated images
- Addressing this challenge by training clinicians or adjustments to the model is necessary to improve usability across varied image qualities and clinical scenarios
- The Python back-end of OASIS, reliance on a structured query language (SQL) database, and recent model integrations contributed to software bloat,
- This led to longer times for opening an image, inferring masks with the model, and the overall editing process
Take home.
These findings suggest that OASIS significantly streamlines the analysis of meibography images, and allows for a more objective and efficient evaluation of MGD.
By implementing deep learning models for gland inference and providing a suite of custom annotation tools, OASIS may reduce the time burden on clinicians while maintaining accuracy.
Plus: OASIS demonstrates the potential clinical utility of AI-driven tools in improving ophthalmic image analysis.