Fundamental Heart Sound Classification using the Continuous Wavelet Transform and Convolutional Neural Networks

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Accurate identification of heart sounds is crucial for diagnosing heart valve pathologies.
• Timing of abnormal heart sounds provides key diagnostic information.
• Heart sound segmentation is essential for developing effective heart sound analysis systems.

Purpose of the Study:

• To investigate the classification of fundamental heart sounds using continuous wavelet transform (CWT) scalograms and convolutional neural networks (CNNs).
• To compare the performance of CNNs, support vector machines (SVM), and k-nearest neighbors (kNN) in classifying heart sounds based on CWT scalograms.
• To evaluate the effectiveness of CNN-extracted features against traditional Linear Binary Pattern (LBP) features.

Main Methods:

• Magnitude scalograms were generated from heart sound samples using the Morse analytic wavelet.
• Convolutional Neural Networks (CNNs) were trained and tested on these scalograms.
• Classification performance was compared between CNNs, SVM, and kNN classifiers.
• Features extracted from CNNs were compared with LBP features.

Main Results:

• The CNN achieved an average classification accuracy of 86% for distinguishing between the first and second heart sounds.
• CNN features classified by SVM yielded a similar accuracy of 85.9%.
• CNN features outperformed LBP features when used with both SVM and kNN classifiers.

Conclusions:

• Continuous Wavelet Transform (CWT) and Convolutional Neural Networks (CNNs) demonstrate significant potential for heart sound analysis.
• This methodology offers a promising approach for accurate classification of fundamental heart sounds.
• The findings support the integration of CWT and CNNs in advanced cardiac diagnostic tools.