Classification of heart murmurs using cepstral features and support vector machines

Area of Science:

• Cardiology
• Biomedical Signal Processing
• Machine Learning

Background:

• Heart murmurs are abnormal heart sounds indicating potential cardiac pathology, such as valve disease.
• Classifying murmurs into systolic and diastolic types is crucial for diagnosis.
• Accurate automated classification of heart sounds can aid clinical decision-making.

Purpose of the Study:

• To investigate the efficacy of cepstral features for classifying heart sounds.
• To compare cepstral features with Short-Term Fourier Transform (STFT) and Discrete Wavelet Transform (DWT) features.
• To evaluate the performance of k-nearest neighbor (kNN), multilayer perceptron (MLP), and support vector machines (SVM) classifiers for heart sound classification.

Main Methods:

• Extracted cepstral features from heart sound signals.
• Utilized kNN, MLP, and SVM classifiers.
• Compared cepstral features against STFT and DWT features using the same classifiers.
• Conducted classification experiments on heart sound samples from various web sources.

Main Results:

• Support Vector Machines (SVM) trained on cepstral features demonstrated the highest classification accuracy.
• Cepstral features combined with SVM achieved approximately 95% accuracy in classifying heart sounds.
• Cepstral features showed superior performance compared to STFT and DWT features across the tested classifiers.

Conclusions:

• Cepstral feature extraction is a promising method for automated heart sound classification.
• SVM classifier offers robust performance when trained with cepstral features for murmur detection.
• This approach holds potential for improving the diagnosis of cardiac pathologies through heart sound analysis.