Faithful, Interpretable Chest X-ray Diagnosis with Anti-Aliased B-cos Networks

2025-07-22

AI Summary

Existing B-cos networks for chest X-ray multi-label diagnosis suffer from high-frequency aliasing artifacts in saliency maps and are inherently limited to single-label classification. Method: We propose an enhanced interpretable B-cos network: (1) FLCPooling and BlurPool are integrated to suppress aliasing and eliminate jagged artifacts in explanation maps; (2) B-cos is extended— for the first time—to multi-label classification via a weight-alignment mechanism that directly produces class-specific, post-processing-free explanations. Results: On public chest X-ray datasets, the model achieves diagnostic performance competitive with state-of-the-art methods while generating high-fidelity, artifact-free saliency maps. It thus simultaneously satisfies clinical requirements for both diagnostic accuracy and interpretability reliability, advancing the practical deployment of explainable deep learning in multi-label medical image diagnosis.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Transparent, Interpretable, Explainable ML
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Responsible Web: Algorithmic accountability and transparency on the web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Faithfulness and interpretability are essential for deploying deep neural networks (DNNs) in safety-critical domains such as medical imaging. B-cos networks offer a promising solution by replacing standard linear layers with a weight-input alignment mechanism, producing inherently interpretable, class-specific explanations without post-hoc methods. While maintaining diagnostic performance competitive with state-of-the-art DNNs, standard B-cos models suffer from severe aliasing artifacts in their explanation maps, making them unsuitable for clinical use where clarity is essential. Additionally, the original B-cos formulation is limited to multi-class settings, whereas chest X-ray analysis often requires multi-label classification due to co-occurring abnormalities. In this work, we address both limitations: (1) we introduce anti-aliasing strategies using FLCPooling (FLC) and BlurPool (BP) to significantly improve explanation quality, and (2) we extend B-cos networks to support multi-label classification. Our experiments on chest X-ray datasets demonstrate that the modified $ ext{B-cos}_ ext{FLC}$ and $ ext{B-cos}_ ext{BP}$ preserve strong predictive performance while providing faithful and artifact-free explanations suitable for clinical application in multi-label settings. Code available at: $href{https://github.com/mkleinma/B-cos-medical-paper}{GitHub repository}$.

Problem

• Improve interpretability of B-cos networks for medical imaging
• Eliminate aliasing artifacts in explanation maps for clinical use
• Extend B-cos networks to multi-label chest X-ray classification

Innovation

• Anti-aliased B-cos networks for clear explanations
• Extended B-cos for multi-label classification
• FLCPooling and BlurPool improve explanation quality