REFUGE2 Challenge: A Treasure Trove for Multi-Dimension Analysis and Evaluation in Glaucoma Screening

TL;DR

REFUGE2 addresss glaucoma screening using color fundus photographs by releasing a multi-device dataset annotated for glaucoma presence, optic disc/cup segmentation, and fovea localization, along with an online evaluation framework. The study reports baseline results from top teams across three tasks, analyzes domain adaptation strategies (e.g., test-time training and adversarial UDA), and discusses how multi-device data influence generalization and clinical applicability. Key findings show strong online performance with domain shifts; OD/OC segmentation can generalize better across devices than fovea localization, and ensemble methods offer selective gains. The work highlights the clinical relevance of robust, device-agnostic analysis and sets the stage for richer multi-modality, demographic-aware benchmarking in glaucoma screening.

Abstract

With the rapid development of artificial intelligence (AI) in medical image processing, deep learning in color fundus photography (CFP) analysis is also evolving. Although there are some open-source, labeled datasets of CFPs in the ophthalmology community, large-scale datasets for screening only have labels of disease categories, and datasets with annotations of fundus structures are usually small in size. In addition, labeling standards are not uniform across datasets, and there is no clear information on the acquisition device. Here we release a multi-annotation, multi-quality, and multi-device color fundus image dataset for glaucoma analysis on an original challenge -- Retinal Fundus Glaucoma Challenge 2nd Edition (REFUGE2). The REFUGE2 dataset contains 2000 color fundus images with annotations of glaucoma classification, optic disc/cup segmentation, as well as fovea localization. Meanwhile, the REFUGE2 challenge sets three sub-tasks of automatic glaucoma diagnosis and fundus structure analysis and provides an online evaluation framework. Based on the characteristics of multi-device and multi-quality data, some methods with strong generalizations are provided in the challenge to make the predictions more robust. This shows that REFUGE2 brings attention to the characteristics of real-world multi-domain data, bridging the gap between scientific research and clinical application.

Figures (21)

• Figure 1: REFUGE2 challenge tasks: glaucoma classification, optic disc/cup segmentation, and fovea localization in CFPs.
• Figure 2: Samples collected from the four camera devices. First row: glaucoma, second row: non-glaucoma.
• Figure 3: t-SNE representations of the original images in the datasets collected from the four camera devices. 0 (red): Zeiss, 1 (blue): Canon, 2 (green): KOWA, 3 (purple): TOPCON.
• Figure 4: Schematic diagram of TTT strategy.
• Figure 5: The framework of the EyeStar team in Task 1. In distributions align module, yellow represents the sample with label 0 and blue represents the sample with label 1; the solid line represents the sample in REFUGE1, and the dotted line represents the sample in SEED.