Multiscale analysis of heart sound signals in the wavelet domain for heart murmur detection

Area of Science:

• Cardiology and Biomedical Signal Processing
• Focuses on the analysis of cardiovascular sounds and the development of diagnostic tools.

Background:

• Heart murmurs, atypical heart sounds from blood flow, are critical indicators of cardiovascular disease.
• Current detection methods for heart murmurs do not fully exploit the information within heart sound signals.
• There is a need for advanced signal processing techniques to improve the accuracy and efficiency of murmur detection.

Purpose of the Study:

• To propose a new set of multiscale features for improved heart murmur detection.
• To leverage wavelet domain analysis to characterize scaling and complexity properties of heart sounds.
• To evaluate the diagnostic performance of these novel features in identifying heart murmurs.

Main Methods:

• Development of multiscale features based on fractal analysis (scaling properties) and wavelet entropy (complexity).
• Characterization of heart sound signals in the wavelet domain to extract these novel features.
• Evaluation of feature diagnostic performance using various classification algorithms, including support vector machines.

Main Results:

• The proposed wavelet-based multiscale features achieved 76.61% accuracy in detecting heart murmurs using a support vector machine classifier.
• Demonstrated competitive performance compared to existing deep learning methods.
• Required significantly fewer features than conventional approaches, indicating high efficiency.

Conclusions:

• Scaling and complexity properties of heart sounds, analyzed via wavelet domain, are potential biomarkers for cardiovascular disease.
• The proposed feature set offers a promising and efficient approach for enhancing the accuracy of heart murmur detection.
• This method provides a valuable alternative for identifying and managing cardiovascular conditions.