An efficient heart murmur recognition and cardiovascular disorders classification system.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Heart murmurs indicate potential cardiovascular disorders.
• Accurate detection and classification are crucial for timely diagnosis.
• Phonocardiogram (PCG) signals offer a non-invasive method for heart sound analysis.

Purpose of the Study:

• To develop a robust system for detecting heart murmurs.
• To classify associated cardiovascular disorders using PCG signals.
• To optimize feature extraction and classification for computational efficiency.

Main Methods:

• Acquired a dataset of PCG signals from 283 volunteers using an electronic stethoscope.
• Extracted 50 Mel-Frequency Cepstrum Coefficients (MFCC) features utilizing systole and diastole intervals.
• Employed iterative backward elimination to reduce feature dimensionality to 26.
• Trained and validated Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) classifiers using cross-validation and data holdout.

Main Results:

• Achieved a classification accuracy of 92.6% using selected MFCC features and a medium Gaussian SVM classifier.
• Identified an optimal MFCC feature vector of dimension 26.
• Demonstrated a good bias-variance trade-off, indicating a well-generalized model.

Conclusions:

• The developed system effectively detects heart murmurs and classifies cardiovascular disorders.
• MFCC features combined with SVM provide a computationally tractable and accurate approach.
• The model shows potential for reliable future predictions in cardiovascular health assessment.