A new HCM heart sound classification method based on weighted bispectrum features.

Abstract

Hypertrophic cardiomyopathy (HCM), including obstructive HCM and non-obstructive HCM, can lead to sudden cardiac arrest in adolescents and athletes. Early diagnosis and treatment through auscultation of different types of HCM can prevent the occurrence of malignant events. However, it is challenging to distinguish the pathological information of HCM related to differential left ventricular outflow tract pressure gradients. To address this issue, a classification method based on weighted bispectrum features of heart sounds (HSs) is proposed for efficient and cost-effective HCM analysis. Preprocessing is first applied to remove background noise during HS acquisition. Then, the bispectrum contour map is calculated, and 56-dimensional features are extracted to represent the pathological information of HCM. Next, an adaptive threshold weighting mutual information method is proposed for feature selection and weighted fusion. Finally, the CNN-RF classifier model is built to automatically identify different types of HCM cases. A clinical dataset of normal and two types of HCM HSs is utilized for validation. The results show that the proposed method performs well, with a classification accuracy reaching 94.4%. It provides a reliable reference for HCM diagnosis in young patients in clinical settings.