Multistage decision-based heart sound delineation method for automated analysis of heart sounds and murmurs

Area of Science:

• Biomedical Engineering
• Cardiovascular Signal Processing
• Artificial Intelligence in Healthcare

Background:

• Accurate delineation of heart sounds and murmurs is crucial for diagnosing cardiac conditions.
• Existing methods often struggle with variable signal amplitudes and complex murmur types.

Purpose of the Study:

• To develop a robust multistage decision-based heart sound delineation (MDHSD) method for automatic identification of cardiac sound features.
• To evaluate the accuracy and robustness of the MDHSD method on diverse phonocardiogram (PCG) signals.

Main Methods:

• The MDHSD method utilizes Gaussian kernels based signal decomposition (GSDs) to separate low and high-frequency components.
• A multistage decision-based delineation (MDBD) algorithm, including envelope extraction and adaptive thresholding, determines fiducial points.
• The GSD algorithm removes low-frequency artifacts and decomposes the signal into LF (S1-S4) and HF (murmurs, HPSs) components.

Main Results:

• The MDHSD method achieved high performance with average sensitivity of 98.22%, positive predictivity of 97.46%, and overall accuracy of 95.78%.
• Delineation errors for sound start and end points were minimal, averaging 4.52 ms and 4.14 ms, respectively.
• The algorithm demonstrated improved accuracy in delineating heart sounds with time-varying amplitudes and various murmur types.

Conclusions:

• The proposed MDHSD method offers a robust and accurate solution for delineating key components of cardiac auscultation signals.
• This advanced delineation technique shows significant potential for integration into automated heart sound and murmur classification systems.
• The MDHSD method enhances the reliability of PCG signal analysis for improved cardiovascular diagnostics.