Butterfly Optimization and Deep Learning to Classify Heart Sound Signal

Heart sound signals, or PCG recordings, provide a valuable non-invasive alternative to traditional diagnostic methods like CXR and blood tests for detecting cardiovascular disease. In this study, PCG data are collected, and features such as MFCC and WST are extracted to capture both time and frequency information. These features are used to classify the signals into normal and abnormal categories using DL models, including CNN, Bi-LSTM, and Bi-RNN. The study conducts extensive experiments using individual and combined features, showing that combining features enhances classification accuracy. The proposed CNN model, optimized with the Butterfly Optimization Algorithm, achieves an impressive classification accuracy of 99.07%, outperforming other models. This result highlights the effectiveness of using PCG signals and advanced DL techniques for accurate cardiovascular disease detection, offering a noninvasive and efficient alternative for early diagnosis.