[Segmentation of heart sound signals based on duration hidden Markov model]

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Accurate heart sound segmentation is crucial for cardiac cycle analysis and classification.
• Existing segmentation methods often rely on electrocardiograms (ECGs), limiting their use in certain clinical settings.
• Developing ECG-independent segmentation algorithms is essential for broader applicability.

Purpose of the Study:

• To propose and evaluate a novel algorithm for heart sound segmentation based on the duration hidden Markov model (DHMM).
• To achieve accurate segmentation of heart sounds (S1, systole, S2, diastole) without requiring ECG signals.
• To demonstrate the robustness and anti-noise performance of the proposed DHMM algorithm.

Main Methods:

• Positional labeling of heart sound samples.
• Autocorrelation estimation for cardiac cycle duration and Gaussian mixture distribution for sample-state duration modeling.
• Optimization of Hidden Markov Model (HMM) and establishment of the DHMM.
• Viterbi algorithm for state tracking to identify S1, systole, S2, and diastole.

Main Results:

• The DHMM algorithm achieved an average F1 score of 0.933, average sensitivity of 0.930, and average accuracy rate of 0.936 on 500 heart sound samples.
• The proposed algorithm demonstrated superior performance compared to other existing methods.
• The algorithm exhibited high robustness and anti-noise capabilities.

Conclusions:

• The DHMM-based algorithm provides an effective and accurate method for heart sound segmentation independent of ECG signals.
• This approach offers a novel solution for feature extraction and analysis of heart sound signals in clinical environments.
• The demonstrated robustness and accuracy suggest significant potential for improving non-invasive cardiac diagnostics.