Performance of Bispectral Features and Cardiac Severity: Application to Phonocardiographic (PCG) Signals

The phonocardiogram (PCG) is a non-invasive bio-sound signal used to identify cardiovascular pathologies and assess their severity. This work follows studies that have demonstrated the value of Higher-Order Spectral Analysis (HOSA) techniques for monitoring cardiac severity. HOSA features were extracted and then selected using Random Forest Feature Importance and SelectKBest methods. They were subsequently integrated into a k-Nearest Neighbor (KNN) classifier. Out of fourteen initial features, four achieved an accuracy of 99.7%, confirming the relevance of this approach for PCG signal analysis.