Detection of Coronary Artery Disease Based on Clinical Phonocardiogram and Multiscale Attention Convolutional...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Phonocardiogram (PCG) signals contain vital information for diagnosing coronary artery disease (CAD).
• Existing machine learning approaches for CAD detection from PCG often suffer from limited clinical data and inefficient feature utilization.
• Current methods necessitate complex, multi-step processing, including manual feature engineering and classifier design.

Purpose of the Study:

• To develop an efficient and automated method for detecting coronary artery disease (CAD) using phonocardiogram (PCG) signals.
• To address the limitations of existing methods regarding dataset size and feature extraction complexity.
• To propose a novel deep learning model, the multiscale attention convolutional compression network (MACCN), for enhanced PCG analysis.

Main Methods:

• A clinical PCG dataset was curated, comprising 102 subjects with CAD and 82 non-CAD subjects.
• A multiscale attention convolutional block (MACB) was designed, incorporating multiscale convolutions and a channel attention module to capture comprehensive and salient heart sound features.
• A novel downsampling block was introduced to minimize feature loss during network compression.
• The MACCN model was developed to integrate automatic feature extraction and classification, eliminating the need for empirical feature selection.

Main Results:

• The MACCN model achieved high classification performance on the clinical PCG dataset.
• Specific performance metrics include an accuracy of 93.43%, sensitivity of 93.44%, precision of 93.48%, and F1 score of 93.42%.
• The model demonstrated effective feature mining from PCG signals for CAD detection.

Conclusions:

• The proposed MACCN model offers an effective and simplified approach for CAD detection using PCG.
• MACCN successfully integrates automatic feature extraction and classification, reducing processing complexity.
• The study highlights the potential of MACCN for robust PCG-based cardiovascular disease diagnosis.