This scientific article is focused on the detection of heart sounds and the study of their age-related characteristics. Heart sounds are one of the most important indicators of cardiac function, and their accurate analysis allows early detection of heart diseases and effective diagnostics. The study examined S1 and S2 tones, as well as murmurs and other pathological sounds, and effective methods for their analysis.

The main focus of the study was on noise reduction (denoising), signal segmentation, and automatic classification processes. Classical machine learning methods (SVM, KNN, Random Forest, Naive Bayes) and deep learning methods (CNN, RNN, LSTM, BiLSTM, autoencoder) were analyzed. The effectiveness of clean and noisy heart sound data was compared, and the strengths and weaknesses of each method were identified.

The results showed that deep learning methods, especially end-to-end approaches, provide high accuracy for segmentation and classification in ambiguous and noisy data. Classical machine learning methods are effective for small and precise datasets, easy to interpret, and require fewer resources. Combining noise reduction, segmentation, and classification processes significantly improves the automatic detection of heart sounds for diagnostic purposes.

The results of this study serve as a scientific basis for improving diagnostic procedures, as well as for the development of medical devices and smart tools for automatic heart sound analysis.

Heart sounds, S1, S2, murmur, age-related characteristics, automatic classification, noise reduction, machine learning, deep learning.

Springer, Heart Sound Analysis: Concepts and Techniques, 2020.

PhysioNet/CinC Challenge Database, 2016.

Liu, C. et al., Deep Learning for Heart Sound Classification, IEEE Access, 2019.

PASCAL Challenge Database, iStethoscope Pro, 2018.

Zhang, X. et al., Transfer Learning in Heart Sound Analysis, Computers in Biology and Medicine, 2021.

CirCor DigiScope Database, 2020.

Shen, Y. et al., Segmenting Heart Sounds Using Deep Neural Networks, Biomedical Signal Processing, 2022.