An Automatic Approach Using ELM Classifier for HFpEF Identification Based on Heart Sound Characteristics

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Heart failure with preserved ejection fraction (HFpEF) presents a complex diagnostic challenge.
• Existing diagnostic methods can be invasive or lack specificity.
• A non-invasive approach for HFpEF diagnosis is highly desirable.

Purpose of the Study:

• To develop and validate a non-invasive diagnostic method for HFpEF using heart sound (HS) characteristics.
• To assess the efficacy of extreme learning machine (ELM) for HFpEF identification based on HS features.

Main Methods:

• Signal preprocessing using improved wavelet denoising.
• Heart sound segmentation and diastolic/systolic duration ratio calculation via hidden semi-Markov models.
• Feature extraction using multifractal detrended fluctuation analysis (MF-DFA).
• HFpEF classification using extreme learning machine (ELM) and support vector machine (SVM).

Main Results:

• Eleven HS features were extracted to differentiate between healthy individuals and HFpEF patients.
• Statistical analysis identified key diagnostic features.
• The ELM model achieved 96.32% accuracy, 95.48% sensitivity, and 97.10% specificity for HFpEF diagnosis.

Conclusions:

• The proposed non-invasive method utilizing heart sound characteristics is effective for diagnosing HFpEF.
• Heart sound analysis offers a promising avenue for improved HFpEF detection.
• The ELM model demonstrates superior performance compared to SVM in this application.