Unsupervised training of keypoint-agnostic descriptors for flexible retinal image registration

TL;DR

The paper tackles the data-label scarcity problem in color fundus image registration by introducing UnConKeD, an unsupervised descriptor learning approach that is agnostic to the keypoint detector. By randomly sampling keypoints from the retinal ROI and training a descriptor network with multi-view augmented data and the FastAP loss, it eliminates reliance on labeled landmarks and enables detector flexibility. Across extensive experiments on the FIRE dataset with diverse detectors, UnConKeD achieves performance on par with or better than supervised counterparts and demonstrates robustness across detectors and keypoint counts. This detector-agnostic, unsupervised framework advances retinal registration by reducing data requirements while remaining competitive with state-of-the-art methods, offering practical benefits for medical imaging where labeled data are limited.

Abstract

Current color fundus image registration approaches are limited, among other things, by the lack of labeled data, which is even more significant in the medical domain, motivating the use of unsupervised learning. Therefore, in this work, we develop a novel unsupervised descriptor learning method that does not rely on keypoint detection. This enables the resulting descriptor network to be agnostic to the keypoint detector used during the registration inference. To validate this approach, we perform an extensive and comprehensive comparison on the reference public retinal image registration dataset. Additionally, we test our method with multiple keypoint detectors of varied nature, even proposing some novel ones. Our results demonstrate that the proposed approach offers accurate registration, not incurring in any performance loss versus supervised methods. Additionally, it demonstrates accurate performance regardless of the keypoint detector used. Thus, this work represents a notable step towards leveraging unsupervised learning in the medical domain.

Figures (2)

• Figure 1: Proposed unsupervised descriptor training
• Figure 2: Results for the different keypoint detectors in FIRE, measured in Weighted Average of Registration Score AUC