New Deep-learning Software Sets Foundation for MGD Biomarker Development

The newly developed OASIS software can automatically assess thousands of clinical meibography images in a matter of minutes, providing objective measurements of gland count, eyelid area, gland loss area, gland area, percentage of gland loss, percentage of gland area and the Pult meiboscale. Researchers hope to create new clinical software tools from the work. These images from the paper show composite eyelid, gland and gland loss masks illustrating different Pult meiboscores from 0 to 4 as determined by OASIS. Eyelid masks are represented by the orange mask. Glands are uniquely labeled with colors between dark blue and light green. Gland color has no additional meaning. The gland loss region is identified by the red mask. Photo: Joseph N, et al. Transl Vis Sci Technol. 2025;14(9):22. Click image to enlarge. While numerous metrics can be obtained from meibography images to help assess meibomian gland dysfunction (MGD), the interpretation of this data is largely subjective and prone to variability. A group of researchers recently proposed a more objective approach to evaluating the meibomian glands: an interactive application called OASIS (Ophthalmic Segmentation and Analysis Software) that combines deep learning-based eyelid and gland segmentation, manual editing tools and quantitative metric generation. A study published recently in Translational Vision Science & Technology determined that the software could quantitatively analyze MGD 87% quicker than manual analysis by human graders.“The OASIS software was developed through a collaboration of researchers and clinicians in the Department of Biomedical Engineering, Case School of Engineering School of Medicine, University Hospitals of Cleveland and our Cornea Image Analysis Reading Center,” comments Thomas Stokkermans, OD, PhD, one of the study’s authors. He notes that the accuracy and efficiency of the new application, as demonstrated by their findings, “will enhance our ability to do research on dry eye, especially to help us understand the natural history and progression of meibomian gland loss and dysfunction.” Moreover, Dr. Stokkermans adds, “It will also help us study how dry eye interventions affect the meibomian glands, including treatments directly targeting the meibomian glands such as heating and expression methods, MG probing, antimicrobial treatments such as tea tree oil, hypochlorous acid and lotilaner.” The study evaluated 2,439 meibography images collected from 325 patients across 11 clinical sites using the LipiView II device. Each participant underwent imaging at two points: an initial visit and a follow-up 90 days later. In the manual analysis, clinicians annotated three distinct masks for each image: one for the eyelid, one for the glands and one for gland loss. The deep learning model incorporated in OASIS facilitated the inference of gland masks, significantly reducing the time necessary for gland-by-gland assessments.When using OASIS, clinicians were able to quantitatively analyze images in less than three minutes, while traditional methods required an average of 15 to 20 minutes of manual work. The software’s accuracy was affirmed through a high level of agreement with human assessments, evidenced by a Cohen's kappa score of 0.79 when comparing manually determined Pult meiboscores to those calculated by OASIS.Dr. Stokkermans explains that while OASIS is currently being used as a research tool for blepharitis and dry eye, its developers “intend to use the OASIS software in dry eye and meibomian gland dysfunction studies in the future” and have a pending patent for research purposes that may also be applicable in clinical settings and meibography tools. However, he notes that it could take several years for new studies to be completed and for the patent to be approved.“This AI-driven meibomian gland analysis tool and others like it can, and are, being adopted into different manufacturers’ meibographers,” says Dr. Stokkermans. “The accuracy, reproducibility, adaptability and ease of interpretation of meibography results will determine the ease of use of our next-generation meibographers. The visuals displayed by the OASIS software are easy to interpret by clinicians and patients alike and will help with education of patients regarding the health of their meibomian glands.”While bringing such insights to clinical practice through better meibography software will take time, the research did yield some dividends today. “I learned that obtaining reproducible results can be challenging due to interfering light reflections, focusing issues and incomplete eversion of the eyelids,” Dr. Stokkermans remarks. “This drove home for me that AI can not solve all of our problems, and we still need good clinicians to operate meibographers so that data can be interpreted accurately and monitored over time by our AI driven tools.”Dr. Stokkermans says the research project also taught him that “the emphasis of all meibography interpretations is on meibomian gland loss, noted using the five-point Pult scale, or a percentage loss, but that there are many other morphological changes that develop in meibomian gland dysfunction.” Morphological changes occurring in MGD that he and the team hope will eventually also be incorporated into an AI-driven tool include the following:Abnormal gap: Unusual spacing between glands.Distorted glands: Meibomian glands with an irregular or misshapen form.Dropout: Atrophy or loss of glandular tissue.Fluffy areas: Amorphous, white-appearing areas where glands should be present.Ghost glands: Glands that are visible but have a faint, transparent appearance.Hooked glands: Glands with a bent or hooked shape.No extension to lid margin: Glands that do not extend to the edge of the eyelid.Overlapping: Glands that appear to overlap each other.Shortened: Glands that are shorter than a healthy gland.Tadpoling: Glands with a "tadpole" shape, having a large head and a thin tail.Thickened: Glands that have thickened, irregular boundaries.Thinned: Glands with thin, threadlike structures.Tortuous glands: Glands that have a winding or twisted shape. The study authors concluded their paper by noting that “future work can use this software and analysis pipeline to automatically assess thousands of clinical meibography images nearly instantly, providing objective evaluations of MG health.”Click here for the journal source. Joseph N, Shivade V, Chen J, et al. Ophthalmic segmentation and analysis software (OASIS): a comprehensive tool for quantitative evaluation of meibography images. Transl Vis Sci Technol. 2025;14(9):22. This article was developed by the editorial staff in conjunction with experts in the field. In the process, AI may have been among the editorial tools used to meet the goals of human editors, who approved all content.