A multi-scale and multi-domain heart sound feature-based machine learning model for ACC/AHA heart failure stage...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Chronic heart failure (CHF) significantly impacts cardiac mechanical activity, necessitating accurate staging for effective management.
• The ACC/AHA heart failure (HF) classification system is crucial for clinical decision-making in CHF patients.
• Objective assessment of CHF severity is challenging, highlighting the need for advanced diagnostic tools.

Purpose of the Study:

• To develop and evaluate a machine learning model for objective ACC/AHA HF stage classification using multi-scale and multi-domain heart sound features.
• To investigate the efficacy of phonocardiogram (PCG) signal analysis for noninvasive CHF staging.
• To compare the performance of different machine learning classifiers for HF staging.

Main Methods:

• A dataset of PCG signals from 275 subjects was analyzed.
• Multi-scale and multi-domain heart sound features were extracted using complementary ensemble empirical mode decomposition and tunable-Q wavelet transform.
• Features were selected using the least absolute shrinkage and selection operator (LASSO) and fed into LS-SVM, DBN, and RF classifiers.

Main Results:

• The LS-SVM model, utilizing combined multi-scale and multi-domain features, outperformed DBN and RF classifiers.
• The LS-SVM model achieved an average sensitivity of 0.821, specificity of 0.955, and accuracy of 0.820 on the testing set.
• This approach demonstrated superior classification performance compared to models using single-domain or single-scale features.

Conclusions:

• Phonocardiogram (PCG) signal analysis offers valuable insights into CHF severity.
• The proposed machine learning model provides a promising noninvasive method for ACC/AHA HF stage classification.
• Objective, feature-rich analysis of heart sounds can significantly aid in managing chronic heart failure.