Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Empowering Multimodal Respiratory Sound Classification with Counterfactual Adversarial Debiasing for Out-of-Distribution Robustness

Heejoon Koo, Miika Toikkanen, Yoon Tae Kim, Soo Yong Kim, June-Woo Kim

TL;DR

The paper tackles spurious metadata-driven biases in multimodal respiratory sound classification by introducing BTS-CARD, a framework that fuses counterfactual debiasing, adversarial debiasing, and counterfactual metadata augmentation within a causal-inference paradigm. By decomposing causal effects into the natural direct effect and total indirect effect, the method learns metadata-insensitive representations while preserving informative pathways, achieving improved in-distribution and out-of-distribution robustness. Empirical results on IND (ICBHI) and OOD (SPRSound) datasets show consistent gains over strong baselines, with ablations confirming the critical role of each component. The approach offers practical benefits for deploying robust respiratory screening across diverse clinical environments without extensive hyperparameter tuning.

Abstract

Multimodal respiratory sound classification offers promise for early pulmonary disease detection by integrating bioacoustic signals with patient metadata. Nevertheless, current approaches remain vulnerable to spurious correlations from attributes such as age, sex, or acquisition device, which hinder their generalization, especially under distribution shifts across clinical sites. To this end, we propose a counterfactual adversarial debiasing framework. First, we employ a causal graph-based counterfactual debiasing methodology to suppress non-causal dependencies from patient metadata. Second, we introduce adversarial debiasing to learn metadata-insensitive representations and reduce metadata-specific biases. Third, we design counterfactual metadata augmentation to mitigate spurious correlations further and strengthen metadata-invariant representations. By doing so, our method consistently outperforms strong baselines in evaluations under both in-distribution and distribution shifts. Code is available at https://github.com/RSC-Toolkit/BTS-CARD.

Empowering Multimodal Respiratory Sound Classification with Counterfactual Adversarial Debiasing for Out-of-Distribution Robustness

TL;DR

The paper tackles spurious metadata-driven biases in multimodal respiratory sound classification by introducing BTS-CARD, a framework that fuses counterfactual debiasing, adversarial debiasing, and counterfactual metadata augmentation within a causal-inference paradigm. By decomposing causal effects into the natural direct effect and total indirect effect, the method learns metadata-insensitive representations while preserving informative pathways, achieving improved in-distribution and out-of-distribution robustness. Empirical results on IND (ICBHI) and OOD (SPRSound) datasets show consistent gains over strong baselines, with ablations confirming the critical role of each component. The approach offers practical benefits for deploying robust respiratory screening across diverse clinical environments without extensive hyperparameter tuning.

Abstract

Multimodal respiratory sound classification offers promise for early pulmonary disease detection by integrating bioacoustic signals with patient metadata. Nevertheless, current approaches remain vulnerable to spurious correlations from attributes such as age, sex, or acquisition device, which hinder their generalization, especially under distribution shifts across clinical sites. To this end, we propose a counterfactual adversarial debiasing framework. First, we employ a causal graph-based counterfactual debiasing methodology to suppress non-causal dependencies from patient metadata. Second, we introduce adversarial debiasing to learn metadata-insensitive representations and reduce metadata-specific biases. Third, we design counterfactual metadata augmentation to mitigate spurious correlations further and strengthen metadata-invariant representations. By doing so, our method consistently outperforms strong baselines in evaluations under both in-distribution and distribution shifts. Code is available at https://github.com/RSC-Toolkit/BTS-CARD.
Paper Structure (13 sections, 11 equations, 3 figures, 4 tables)

This paper contains 13 sections, 11 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodologies
  3. Preliminaries on Causal Inference
  4. Task Formulation
  5. Counterfactual Adversarial Debiasing Framework
  6. Experiments
  7. Experimental Setup
  8. Experimental Results
  9. Main Results
  10. Ablation Studies
  11. Comparative Studies
  12. Parameter Analysis
  13. Conclusion

Figures (3)

  • Figure 1: An Overview of Our Counterfactual Adversarial Debiasing Framework for Multimodal RSC, BTS-CARD.
  • Figure 2: Causal Graph.
  • Figure 3: Parameter analysis across different values of $\alpha$ on IND (ICBHI) and OOD (SPRSound) settings.