Heart Sound Analysis and Murmur Classification using Bark Scale Modulation Spectrogram Features

Detecting heart murmurs is critical for diagnosing cardiovascular conditions, particularly pediatric patients. This study presents a comprehensive approach for processing heart sound recordings and murmur classification using Convolutional Recurrent Neural Networks (CRNN). A methodology is proposed for feature extraction based on Bark scale modulation spectrograms to enhance classifier efficiency in pediatric heart sound analysis. Modulation spectrogram features of various heart sound signals are analyzed, and their efficiency across different categories is evaluated. Statistical and spectral features are extracted, and correlations between normal and abnormal heart sound categories are studied. The Bark scale modulation spectrograms are utilized to train and evaluate CRNN performance in classifying heart sounds as normal or abnormal. Performance evaluation shows a 92% accuracy in murmur classification with lower computational complexity. This study highlights the efficacy of modulation spectrogram-based feature extraction and advanced classification techniques in the automated detection and classification of heart murmurs, marking a significant advancement in non-invasive cardiovascular diagnostics for pediatric patients.