Heart sound classification from unsegmented phonocardiograms

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Automated phonocardiogram (PCG) analysis typically requires segmentation of heart sounds.
• Segmentation is a complex step in automated PCG analysis.
• Accurate classification of heart sounds is crucial for diagnosing cardiac conditions.

Purpose of the Study:

• To assess the feasibility of accurate heart sound classification using short, unsegmented PCG recordings.
• To evaluate the effectiveness of wavelet entropy and spectral amplitude for heart sound classification.
• To compare classification performance with and without signal segmentation.

Main Methods:

• Analysis of 5-second PCG segments from the PhysioNet/Computing in Cardiology Challenge database.
• Calculation of normalized spectral amplitude using Fast Fourier Transform.
• Determination of wavelet entropy using wavelet analysis.
• Implementation of threshold-based classifiers and a classification tree.
• Comparison of results using initial segments (seg 1) and noise-free segments (seg 2).

Main Results:

• Significant differences in wavelet entropy and spectral amplitude were found between normal and abnormal recordings.
• Abnormal recordings showed reduced high-frequency wavelet entropy and increased low-frequency spectral amplitude.
• Classification accuracy using wavelet entropy reached 76%, improving to 80% with noise-free segments.
• A classification tree combining features achieved 79% accuracy.

Conclusions:

• Accurate heart sound classification is feasible without signal segmentation.
• The proposed method offers comparable accuracy to existing algorithms but with reduced complexity.
• This approach simplifies automated PCG analysis.