Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Noise-Robust Contrastive Learning with an MFCC-Conformer For Coronary Artery Disease Detection

Milan Marocchi, Matthew Fynn, Yue Rong

TL;DR

This study tackles CAD detection from PCGs recorded in noisy clinical environments. It introduces a multichannel, energy-based noisy-segment rejection step paired with an MFCC-augmented Conformer encoder trained with a hybrid supervised contrastive loss to enhance noise robustness and class balance. The approach demonstrates improved subject-level accuracy, balanced performance between true and false positive rates, and a smaller model footprint compared with prior methods. The work advances practical CAD screening from PCG data by addressing real-world noise and leveraging multichannel MFCC features within a contrastive learning framework.

Abstract

Cardiovascular diseases (CVD) are the leading cause of death worldwide, with coronary artery disease (CAD) comprising the largest subcategory of CVDs. Recently, there has been increased focus on detecting CAD using phonocardiogram (PCG) signals, with high success in clinical environments with low noise and optimal sensor placement. Multichannel techniques have been found to be more robust to noise; however, achieving robust performance on real-world data remains a challenge. This work utilises a novel multichannel energy-based noisy-segment rejection algorithm, using heart and noise-reference microphones, to discard audio segments with large amounts of nonstationary noise before training a deep learning classifier. This conformer-based classifier takes mel-frequency cepstral coefficients (MFCCs) from multiple channels, further helping improve the model's noise robustness. The proposed method achieved 78.4% accuracy and 78.2% balanced accuracy on 297 subjects, representing improvements of 4.1% and 4.3%, respectively, compared to training without noisy-segment rejection.

Noise-Robust Contrastive Learning with an MFCC-Conformer For Coronary Artery Disease Detection

TL;DR

This study tackles CAD detection from PCGs recorded in noisy clinical environments. It introduces a multichannel, energy-based noisy-segment rejection step paired with an MFCC-augmented Conformer encoder trained with a hybrid supervised contrastive loss to enhance noise robustness and class balance. The approach demonstrates improved subject-level accuracy, balanced performance between true and false positive rates, and a smaller model footprint compared with prior methods. The work advances practical CAD screening from PCG data by addressing real-world noise and leveraging multichannel MFCC features within a contrastive learning framework.

Abstract

Cardiovascular diseases (CVD) are the leading cause of death worldwide, with coronary artery disease (CAD) comprising the largest subcategory of CVDs. Recently, there has been increased focus on detecting CAD using phonocardiogram (PCG) signals, with high success in clinical environments with low noise and optimal sensor placement. Multichannel techniques have been found to be more robust to noise; however, achieving robust performance on real-world data remains a challenge. This work utilises a novel multichannel energy-based noisy-segment rejection algorithm, using heart and noise-reference microphones, to discard audio segments with large amounts of nonstationary noise before training a deep learning classifier. This conformer-based classifier takes mel-frequency cepstral coefficients (MFCCs) from multiple channels, further helping improve the model's noise robustness. The proposed method achieved 78.4% accuracy and 78.2% balanced accuracy on 297 subjects, representing improvements of 4.1% and 4.3%, respectively, compared to training without noisy-segment rejection.
Paper Structure (18 sections, 5 equations, 1 figure, 2 tables, 1 algorithm)

This paper contains 18 sections, 5 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. MATERIALS
  3. Data Aquistion
  4. Dataset
  5. METHOD
  6. Preprocessing
  7. Noisy Segment Rejection
  8. Segmentation
  9. Feature Extraction
  10. Models
  11. Contrastive Learning
  12. Hybrid-Contrastive Loss
  13. Supervised Contrastive Loss
  14. Center Loss
  15. Model Training
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Zoomed-in HM (top) and NM (bottom) signals with noise-corrupted segments highlighted in red. These segments were discarded from all channels during downstream training and inference.