DualStreamFoveaNet: A Dual Stream Fusion Architecture with Anatomical Awareness for Robust Fovea Localization

TL;DR

This paper introduces DualStreamFoveaNet, a transformer-based architecture that fuses multi-cue retinal features with anatomical awareness, significantly improving the robustness and accuracy of fovea localization in diverse retinal images.

Contribution

It presents a novel dual-stream transformer architecture with spatial attention for anatomical feature fusion, enhancing robustness and generalization in fovea localization tasks.

Findings

Achieves state-of-the-art results on multiple datasets.

More robust on diseased and normal retinal images.

Better generalization across datasets.

Abstract

Accurate fovea localization is essential for analyzing retinal diseases to prevent irreversible vision loss. While current deep learning-based methods outperform traditional ones, they still face challenges such as the lack of local anatomical landmarks around the fovea, the inability to robustly handle diseased retinal images, and the variations in image conditions. In this paper, we propose a novel transformer-based architecture called DualStreamFoveaNet (DSFN) for multi-cue fusion. This architecture explicitly incorporates long-range connections and global features using retina and vessel distributions for robust fovea localization. We introduce a spatial attention mechanism in the dual-stream encoder to extract and fuse self-learned anatomical information, focusing more on features distributed along blood vessels and significantly reducing computational costs by decreasing token numbers. Our extensive experiments show that the proposed architecture achieves state-of-the-art performance on two public datasets and one large-scale private dataset. Furthermore, we demonstrate that the DSFN is more robust on both normal and diseased retina images and has better generalization capacity in cross-dataset experiments.