MCOA: A Comprehensive Multimodal Dataset for Advancing Deep Learning in Corneal Opacity Assessment

Abstract

Abstract Corneal opacity remains a major global cause of vision impairment. Its severity is typically assessed subjectively by clinicians using slit lamp examinations of the anterior segment. While anterior segment optical coherence tomography (AS-OCT) provides high-resolution cross-sectional images of the cornea, capturing subtle structural changes, the combination of AS-OCT images with anterior segment photographs delivers a more comprehensive view of the cornea. However, the absence of large-scale, high-quality datasets hinders the development of deep learning algorithms for this purpose. To bridge this gap, we established the most extensive corneal opacity dataset available. The dataset included a total of 6,272 AS-OCT images and 392 corresponding anterior segment photographs. Each image of patients with corneal opacity was carefully annotated to include detailed cornea and corneal opacity information. This robust dataset represented a significant step forward in leveraging deep learning for corneal opacity recognition, empowering AI-driven clinical decision-making and facilitating the creation of personalized treatment plans for affected patients.