Anatomy-Slot: Unsupervised Anatomical Factorization for Homologous Bilateral Reasoning in Retinal Diagnosis

TL;DR

Anatomy-Slot introduces an unsupervised anatomical factorization method that enhances bilateral retinal diagnosis by explicitly modeling structural correspondence across eyes, improving diagnostic accuracy and robustness.

Contribution

The paper proposes Anatomy-Slot, a novel unsupervised approach that decomposes image patches into slots and aligns them across eyes to improve bilateral reasoning in retinal diagnosis.

Findings

Improves AUC by 4.2% over baseline on ODIR-5K dataset.

Provides controlled tests of anatomical correspondence under noise.

Achieves quantitative optic disc grounding and localization analysis.

Abstract

Retinal diagnosis is inherently bilateral: clinicians compare homologous structures across eyes (e.g., optic disc asymmetry), yet most deep models operate on monocular representations. We investigate whether explicit structural correspondence improves diagnosis, and propose Anatomy-Slot to operationalize this hypothesis. Anatomy-Slot introduces an unsupervised anatomical bottleneck by decomposing patch tokens into slots and aligning slots across eyes via bidirectional cross-attention. On ODIR-5K with $n=10$ seeds, the method improves AUC by 4.2% over a matched ViT-L baseline (95% CIs; Wilcoxon signed-rank test, $W=0$, $p=0.002$). Pairing disruption and stress testing under Gaussian noise provide controlled tests of correspondence dependence and robustness under corruption. We further report quantitative optic disc grounding on REFUGE and cross-attention localization analysis.