Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PC-MCL: Patient-Consistent Multi-Cycle Learning with multi-label bias correction for respiratory sound classification

Seung Gyu Jeong, Seong-Eun Kim

TL;DR

PC-MCL tackles two key issues in respiratory sound classification: loss of temporal context across cycles and patient-specific overfitting. It introduces multi-cycle concatenation with a 3-label formulation (normal, crackle, wheeze) to preserve information in mixed samples, plus a patient-matching auxiliary task as a regularizer. The approach yields state-of-the-art ICBHI 2017 performance (65.37% Score) by combining three components that synergistically enhance sensitivity, generalization, and threshold-free discriminability. This framework advances non-invasive, robust pulmonary disease screening by better exploiting temporal structure and patient diversity in respiratory sounds.

Abstract

Automated respiratory sound classification supports the diagnosis of pulmonary diseases. However, many deep models still rely on cycle-level analysis and suffer from patient-specific overfitting. We propose PC-MCL (Patient-Consistent Multi-Cycle Learning) to address these limitations by utilizing three key components: multi-cycle concatenation, a 3-label formulation, and a patient-matching auxiliary task. Our work resolves a multi-label distributional bias in respiratory sound classification, a critical issue inherent to applying multi-cycle concatenation with the conventional 2-label formulation (crackle, wheeze). This bias manifests as a systematic loss of normal signal information when normal and abnormal cycles are combined. Our proposed 3-label formulation (normal, crackle, wheeze) corrects this by preserving information from all constituent cycles in mixed samples. Furthermore, the patient-matching auxiliary task acts as a multi-task regularizer, encouraging the model to learn more robust features and improving generalization. On the ICBHI 2017 benchmark, PC-MCL achieves an ICBHI Score of 65.37%, outperforming existing baselines. Ablation studies confirm that all three components are essential, working synergistically to improve the detection of abnormal respiratory events.

PC-MCL: Patient-Consistent Multi-Cycle Learning with multi-label bias correction for respiratory sound classification

TL;DR

PC-MCL tackles two key issues in respiratory sound classification: loss of temporal context across cycles and patient-specific overfitting. It introduces multi-cycle concatenation with a 3-label formulation (normal, crackle, wheeze) to preserve information in mixed samples, plus a patient-matching auxiliary task as a regularizer. The approach yields state-of-the-art ICBHI 2017 performance (65.37% Score) by combining three components that synergistically enhance sensitivity, generalization, and threshold-free discriminability. This framework advances non-invasive, robust pulmonary disease screening by better exploiting temporal structure and patient diversity in respiratory sounds.

Abstract

Automated respiratory sound classification supports the diagnosis of pulmonary diseases. However, many deep models still rely on cycle-level analysis and suffer from patient-specific overfitting. We propose PC-MCL (Patient-Consistent Multi-Cycle Learning) to address these limitations by utilizing three key components: multi-cycle concatenation, a 3-label formulation, and a patient-matching auxiliary task. Our work resolves a multi-label distributional bias in respiratory sound classification, a critical issue inherent to applying multi-cycle concatenation with the conventional 2-label formulation (crackle, wheeze). This bias manifests as a systematic loss of normal signal information when normal and abnormal cycles are combined. Our proposed 3-label formulation (normal, crackle, wheeze) corrects this by preserving information from all constituent cycles in mixed samples. Furthermore, the patient-matching auxiliary task acts as a multi-task regularizer, encouraging the model to learn more robust features and improving generalization. On the ICBHI 2017 benchmark, PC-MCL achieves an ICBHI Score of 65.37%, outperforming existing baselines. Ablation studies confirm that all three components are essential, working synergistically to improve the detection of abnormal respiratory events.
Paper Structure (15 sections, 2 equations, 3 figures, 5 tables)

This paper contains 15 sections, 2 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Multi-Cycle Concatenation and Bias Correction
  4. Regularization via a Patient-Matching Auxiliary Task
  5. Evaluation and Inference Strategy
  6. Experiments and Results
  7. Experimental Setup
  8. Main Results and Comparison
  9. Ablation Studies
  10. Component Analysis
  11. Analysis of Label Formulation
  12. Auxiliary Task - Negative Sampling Strategy
  13. Threshold-Free Performance Analysis
  14. Conclusion
  15. Acknowledgment

Figures (3)

  • Figure 1: Overview of the proposed PC-MCL framework. Individual respiratory cycles are combined via Multi-Cycle Concatenation, converted to a spectrogram, and fed into a shared backbone encoder. The resulting features are passed to two independent heads for the main pathology classification and the auxiliary patient-matching task.
  • Figure 2: Comparison of feature spaces learned by the conventional 2-label and our proposed 3-label models, visualized using t-SNE. The points represent three types of augmented samples: Pure Normal (green), Pure Abnormal (red), and Mixed Normal+Abnormal (blue).
  • Figure 3: Precision-Recall Curves for the three classes (Normal, Crackle, Wheeze) on the ICBHI test set. The curves compare the performance of the baseline CE model, an intermediate model with concatenation and multi-labeling (Multilabel), and our final proposed model (Ours). Average Precision (AP) scores, averaged over five runs, are shown in the legend for each model.