Heart sound classification based on convolutional neural network with convolutional block attention module.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Cardiovascular diseases (CVDs) are a primary cause of global mortality.
• Accurate and efficient diagnostic tools for CVDs are crucial.
• Current diagnostic methods may lack the precision needed for early detection.

Purpose of the Study:

• To develop an enhanced heart sound classification framework using deep learning.
• To integrate the Convolutional Block Attention Module (CBAM) with a Convolutional Neural Network (CNN).
• To evaluate the performance of the attention-based CNN for classifying heart sounds.

Main Methods:

• Utilized heart sound recordings from the PhysioNet CinC 2016 dataset.
• Processed audio data into spectrograms for analysis.
• Systematically evaluated twelve CNN models with varying CBAM configurations.

Main Results:

• The optimal CNN model with CBAM integration achieved 98.66% accuracy on the primary dataset.
• Validation on an independent PhysioNet 2022 dataset yielded 95.6% accuracy and 96.29% AUC.
• T-SNE visualizations demonstrated clear class separation, indicating effective feature extraction.

Conclusions:

• Selective integration of CBAM significantly improves CNN performance in heart sound classification.
• Attention-based architectures are effective for medical signal classification.
• The developed framework shows potential for real-world clinical application in diagnosing CVDs.