Cardiac sound murmurs classification with autoregressive spectral analysis and multi-support vector machine technique

Area of Science:

• Cardiology and Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Accurate classification of cardiac murmurs is crucial for diagnosing heart valvular disorders.
• Traditional methods may lack the precision needed for subtle murmur differentiation.
• Novel signal processing and machine learning approaches are needed for enhanced cardiac sound analysis.

Purpose of the Study:

• To propose and validate a novel cardiac sound spectral analysis method for classifying heart murmurs.
• To introduce new diagnostic features (Fmax and Fwidth) derived from the normalized autoregressive power spectral density (NAR-PSD) curve.
• To employ multi-support vector machine (SVM) classifiers for detecting cardiac abnormality and discriminating between different heart valvular disorders.

Main Methods:

• Acquisition of 489 cardiac sound signals (196 normal, 293 abnormal) from healthy subjects and patients with specific valvular disorders.
• Calculation of normalized autoregressive power spectral density (NAR-PSD) and extraction of Fmax and Fwidth features.
• Implementation of multi-SVM classifiers, comprising six SVM modules, for classifying normal and abnormal heart sounds.

Main Results:

• The proposed method achieved high classification accuracies, ranging from 71-98.9% for specific disorder classifications.
• Overall classification of normal versus abnormal sounds demonstrated excellent performance with 99.9% specificity and 99.5% sensitivity.
• The effectiveness of the SVM modules was significantly influenced by the chosen threshold value (THV).

Conclusions:

• The novel cardiac sound spectral analysis method using NAR-PSD and SVM classifiers is highly effective for classifying heart murmurs.
• The proposed Fmax and Fwidth features provide valuable diagnostic information for cardiac sound interpretation.
• This approach offers a promising, efficient, and validated tool for the detection and classification of heart valvular disorders.