Robust classification of heart valve sound based on adaptive EMD and feature fusion

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Cardiovascular disease (CVD) is a leading global cause of mortality.
• Diagnosing CVD through heart sound analysis is an active research area.
• Distinguishing heart valve defects from normal heart sounds requires sophisticated analysis techniques.

Purpose of the Study:

• To develop and validate a novel method for accurate heart sound classification.
• To improve the diagnosis of heart valve defects using adaptive empirical mode decomposition (EMD) and feature fusion.
• To enhance the accuracy and robustness of heart sound analysis for cardiovascular disease detection.

Main Methods:

• Adaptive empirical mode decomposition (EMD) was employed to screen intrinsic mode function (IMF) components.
• Hausdorff Distance was used for adaptive thresholding to select IMF components for signal reconstruction.
• Multidimensional features from time, frequency, and nonlinear domains were extracted and fused with four cardiac reserve time features.
• Feature selection algorithms ranked fused features, and three classifiers (random forest, decision tree, K-nearest neighbor) were trained.

Main Results:

• The proposed method achieved a classification accuracy of 99.3%, a 1.9% improvement over previous methods.
• Extensive experimental evaluations demonstrated the method's effectiveness on both open-source and custom databases.
• The fusion features and the proposed method showed excellent robustness and effectiveness in heart sound classification.

Conclusions:

• The developed method significantly enhances the accuracy of heart sound classification for cardiovascular disease diagnosis.
• Adaptive EMD and feature fusion techniques are effective in identifying subtle patterns indicative of heart valve defects.
• The findings support the clinical utility of advanced signal processing techniques for non-invasive CVD screening.