Segmentation and classification of heart sounds

Cardiovascular diseases are significant health issues worldwide, and early diagnosis can improve patients' quality of life and enhance treatment processes. In this study, a method for in – depth analysis of heart sounds and diagnosis of cardiovascular diseases has been developed. A comprehensive analysis was conducted on two different datasets of heart sounds. Initially, the datasets underwent preprocessing steps including notch filtering, elliptic filtering, and normalization. Subsequently, peak values were identified using the Otsu thresholding method, followed by segmentation using Shannon energy. Feature extraction methods such as Empirical Mode Decomposition (EMD) and Mel Frequency Cepstral Coefficients (MFCC) were applied to the obtained segments. These features facilitated the representation of heart sounds in the time – frequency domain. In the final step, various classification algorithms (k – Nearest Neighbors, Support Vector Machines, Artificial Neural Networks, Convolutional Neural Networks) were employed to classify diseased and healthy heart sounds. Experimental results demonstrate that the proposed method achieves high accuracy rates and reliable outcomes. This study presents an effective tool for early diagnosis of cardiovascular diseases and monitoring treatment processes.