C$^2$A: Coupling Spatial Evidence with Clinical Priors via Co-occurrence Aware Class Attention for Multi-Label Chest X-Ray Classification

2026-08-10

AI Summary

This work addresses the limitations of existing methods in multi-label chest X-ray classification, which often neglect the spatial localization of lesions and disease co-occurrence patterns, thereby underperforming on highly co-occurring conditions with ambiguous spatial evidence. To overcome this, the authors propose the C²A classification head, which for the first time explicitly integrates clinical label co-occurrence priors into a class-aware spatial attention mechanism via learnable bilinear interactions. Specifically, it generates disease-specific spatial descriptors through class attention maps and constructs a graph based on co-occurrence statistics, performing a single step of residual message passing to jointly leverage spatial evidence and co-occurrence priors. With negligible additional computational cost, the method achieves a macro-averaged AUROC of 0.895 on CheXpert, significantly outperforming state-of-the-art baselines—particularly improving performance on high-co-occurrence classes such as atelectasis by up to +1.5 AUROC points.

Abstract

Thoracic pathologies rarely occur in isolation, yet standard multi-label classifiers rely on shared global descriptors, discarding \emph{where} findings lie and \emph{how} they co-occur. We propose \textbf{C$\mathbf{^2}$A} (Co-occurrence Aware Class Attention), a classification head that explicitly couples spatial evidence with clinical priors. First, C$^2$A casts pooling as an expectation over learned per-class spatial attention maps, yielding localized descriptors for each disease. Second, it couples these descriptors via a learnable graph warm-started from empirical label co-occurrence. A single residual message-passing step shares evidence among related findings, proving to be a bounded perturbation of the identity where co-occurrence enters each logit through an explicit bilinear interaction. On CheXpert, C$^2$A achieves a superior $0.895$ macro-mean AUROC, outperforming advanced context-gating baselines. Crucially, gains concentrate on highly co-occurrent classes with ambiguous spatial evidence (rescuing Atelectasis by $+1.5$ over GCG), demonstrating the prior's regularizing effect with a negligible overhead of one linear projection and a $C\!\times\!C$ edge matrix.

Problem

• multi-label classification
• chest X-ray
• spatial localization
• label co-occurrence
• thoracic pathologies

Innovation

• co-occurrence aware attention
• spatial evidence
• clinical priors
• multi-label chest X-ray classification
• graph-based message passing