Entropy-Based Phonocardiogram Classification Using Continuous and Synchrosqueezed Wavelet Transforms: A Systematic...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Conventional cardiac auscultation faces limitations due to examiner variability and sensitivity.
• Automatic phonocardiogram (PCG) classification requires robust methods to handle signal non-stationarity and extract discriminative features.

Purpose of the Study:

• To develop a computational framework integrating wavelet analysis with entropy-based features for distinguishing normal and abnormal heart sounds.
• To evaluate the performance of this framework for standardized cardiac screening.

Main Methods:

• Utilized the PhysioNet Computing in Cardiology Challenge 2016 database.
• Applied continuous wavelet transform (CWT) and various entropy measures (Shannon, Rényi, Tsallis, spectral, permutation, sample) across different frequency bands.
• Employed a regularized multi-layer perceptron with dropout for classification and 5-fold cross-validation for evaluation.

Main Results:

• Achieved an area under the receiver operating characteristic curve (AUROC) of 0.972 and balanced accuracy of 0.915.
• Standard CWT outperformed synchrosqueezed CWT, and band-specific entropy analysis significantly contributed to performance.
• Demonstrated 96% sensitivity at 90% specificity.

Conclusions:

• The integrated wavelet analysis and entropy engineering approach achieves state-of-the-art performance in heart sound classification.
• The methodology is computationally efficient (1.2s) and interpretable.
• Offers practical potential for point-of-care cardiac screening, especially in resource-limited settings.