Joseph, Naomi, Shivade, Ved, Chen, Jiawei, Marshall, Ian, Jennings, Dominique, Menegay, Harry, Ramamirtham, Ramkumar, Wilson, david, Benetz, Beth-Ann and Stokkermans, Thomas (2025) Ophthalmic Segmentation and Analysis Software (OASIS): A Comprehensive Tool for Quantitative Evaluation of Meibography Images. Translational Vision Science & Technology, 14 (9). pp. 1-11. ISSN ISSN: 2164-2591

Abstract

Meibomian gland dysfunction (MGD) is a prevalent cause of evaporative dry eye disease, resulting from the atrophy and reduced lipid secretion of the meibomian glands (MG). Current methods for evaluating MGs utilizing meibography rely on subjective assessment of gland loss. This study proposes an interactive image editor, Ophthalmic Segmentation and Analysis Software (OASIS), which includes features for manual and semi-automatic (assisted) analysis of meibography images. A natural history study collected 2,439 meibography images from 325 patients, which clinicians subsequently analyzed with OASIS. As part of OASIS’s manual analysis process, clinicians annotate three masks per image: the eyelid, glands, and gland loss. In the assisted process, OASIS can infer the gland mask using integrated deep-learning models, reducing the need for timely gland-by-gland annotation from the user. The Graphical User Interface (GUI) of OASIS also provides enhancement features and analysis tools, allowing users to apply filters, annotate ROIs, and calculate relevant clinical metrics. Metrics computed include gland count, eyelid area, gland loss area, gland area, gland loss percentage, gland area percentage, and, critically, a gland loss score based on the established Pult meiboscale. OASIS overcomes the limitations of previous methods by allowing clinicians to perform quantitative analyses of MGD in under 3 minutes, an 87% reduction in time compared to manual analysis. The software accurately calculates Pult meiboscale grades for meibography images with a fair agreement between the clinician and the software (kappa = 0.79). OASIS leads the efforts to develop in-depth quantitative biomarkers for MGD in clinical practices and therapeutic research.

Keywords: Meibomian gland dysfunction, software, deep learning, gland segmentation

Item Type:	Article
Keywords:	Meibomian gland dysfunction, software, deep learning, gland segmentation
Date Deposited:	27 Sep 2025 00:45
Last Modified:	27 Sep 2025 00:45
URI:	https://oak.novartis.com/id/eprint/54942