Related Experiment Video
Updated: Jul 26, 2025
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
Parameter-Efficient Densely Connected Dual Attention Network for Phonocardiogram Classification
Insights
A novel densely connected dual attention network (DDA) enhances cardiovascular disease diagnosis using phonocardiogram (PCG) data. This efficient deep learning model improves heart sound classification without complex pre-processing.
Area of Science:
- Cardiology
- Biomedical Engineering
- Artificial Intelligence
Background:
- Cardiac auscultation using phonocardiograms (PCG) is a vital non-invasive diagnostic tool for cardiovascular diseases (CVDs).
- Challenges in PCG analysis include inherent murmurs and limited supervised data, hindering accurate heart sound classification.
- Current deep learning methods often require extensive pre-processing, relying on time-consuming expert engineering.
Purpose of the Study:
- To propose a parameter-efficient, densely connected dual attention network (DDA) for automated heart sound classification.
- To develop an end-to-end deep learning architecture that integrates hierarchical feature extraction and attention mechanisms.
- To improve the computational efficiency and classification performance of computer-aided heart sound analysis.
Main Methods:
- A densely connected structure was employed for hierarchical extraction of heart sound features.
- A dual attention mechanism, utilizing self-attention, was implemented to aggregate local and global feature dependencies across positional and channel axes.
- The proposed DDA model was evaluated using stratified 10-fold cross-validation on the Cinc2016 benchmark dataset.
Main Results:
- The DDA model demonstrated superior performance compared to existing 1D deep learning models on the Cinc2016 benchmark.
- The network achieved significant computational efficiency, reducing the reliance on time-consuming pre-processing steps.
- Hierarchical feature extraction and dual attention mechanisms effectively captured complex patterns in heart sound data.
Conclusions:
- The proposed DDA model offers an effective and computationally efficient solution for heart sound classification.
- This approach advances computer-aided diagnosis of cardiovascular diseases by leveraging deep learning and attention mechanisms.
- The DDA network provides a promising direction for developing robust and accessible diagnostic tools in cardiology.
Abstract:
Cardiac auscultation, exhibited by phonocardiogram (PCG), is a non-invasive and low-cost diagnostic method for cardiovascular diseases (CVDs). However, deploying it in practice is quite challenging, due to the inherent murmurs and a limited number of supervised samples in heart sound data. To solve these problems, not only heart sound analysis based on handcrafted features, but also computer-aided heart sound analysis based on deep learning have been extensively studied in recent years. Though with elaborate design, most of these methods still use additional pre-processing to improve classification performance, which heavily relies on time-consuming experienced engineering. In this article, we propose a parameter-efficient densely connected dual attention network (DDA) for heart sound classification. It combines two advantages simultaneously of the purely end-to-end architecture and enriched contextual representations of the self-attention mechanism. Specifically, the densely connected structure can automatically extract the information flow of heart sound features hierarchically. Alongside, improving contextual modeling capabilities, the dual attention mechanism adaptively aggregates local features with global dependencies via a self-attention mechanism, which captures the semantic interdependencies across position and channel axes respectively. Extensive experiments across stratified 10-fold cross-validation strongly evidence that our proposed DDA model surpasses current 1D deep models on the challenging Cinc2016 benchmark with significant computational efficiency.
