Classification of heart sounds based on quality assessment and wavelet scattering transform

Area of Science:

• Cardiology
• Biomedical Signal Processing
• Machine Learning

Background:

• Cardiovascular diseases necessitate accurate diagnostic tools.
• Automatic heart sound classification is crucial for early detection.
• Existing methods may lack robustness in signal quality.

Purpose of the Study:

• To propose an effective heart sound classification method.
• To enhance classification accuracy through quality assessment and advanced feature extraction.
• To evaluate the performance of the proposed method on a public dataset.

Main Methods:

• Heart sound signal quality assessed using Ratio of Zero Crossings (RZC) and Root Mean Square of Successive Differences (RMSSD).
• Wavelet scattering transform employed for feature extraction from selected signal segments.
• Support Vector Machine (SVM) utilized for classification with a voting approach.

Main Results:

• The method achieved 92.23% accuracy, 96.62% sensitivity, and 90.65% specificity on the PhysioNet CinC 2016 dataset.
• Detailed analysis of RZC and RMSSD impact on classification outcomes.
• High classification performance for normal and abnormal heart sound samples.

Conclusions:

• The proposed quality assessment and wavelet scattering transform method effectively classifies heart sounds.
• This technique offers a promising tool for non-invasive cardiovascular disease diagnosis.
• The findings support the utility of advanced signal processing in clinical applications.