Efficient heart sound segmentation and extraction using ensemble empirical mode decomposition and kurtosis features

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Accurate segmentation of heart sounds (S1, S2) is crucial for cardiac diagnosis.
• Existing methods may struggle with noise and accuracy in clinical auscultatory data.
• Automated analysis of phonocardiograms requires robust feature extraction and segmentation algorithms.

Purpose of the Study:

• To develop and validate an efficient heart sound segmentation (HSS) method for automatic detection and extraction of S1 and S2 sounds.
• To evaluate the performance of the proposed HSS-EEMD/K scheme in a real clinical environment with diverse cardiac conditions.
• To assess the noise robustness and compare the accuracy of the HSS-EEMD/K method against other established techniques.

Main Methods:

• The proposed heart sound segmentation (HSS) scheme, HSS-EEMD/K, utilizes ensemble empirical mode decomposition (EEMD) combined with kurtosis features.
• The method analyzes raw heart auscultatory data to locate and extract first (S1) and second (S2) heart sounds.
• Performance was evaluated on 43 recordings from normal subjects and patients with aortic stenosis and mitral regurgitation.

Main Results:

• The HSS-EEMD/K approach achieved 94.56% accuracy in determining heart sound locations and correctly segmented 83.05% of heart cycles.
• Noise stress tests demonstrated the method's robustness against additive Gaussian and respiratory noises.
• Compared to four other methods, HSS-EEMD/K showed superior accuracy (7-19% increase) and prediction power (4-9% increase).

Conclusions:

• The HSS-EEMD/K method offers an efficient and accurate solution for heart sound segmentation.
• Its high performance and noise robustness support its potential for clinical application.
• This technique can enhance the diagnostic value of heart sound analysis in routine clinical practice.