A Comprehensive Survey on Heart Sound Analysis in the Deep Learning Era

Abstract

Heart sound auscultation has been demonstrated to be beneficial in clinical usage for early screening of cardiovascular diseases. Due to the high requirement of well-trained professionals for auscultation, automatic auscultation benefiting from signal processing and machine learning can help auxiliary diagnosis and reduce the burdens of training professional clinicians. Nevertheless, classic machine learning is limited to performance improvement in the era of big data. Deep learning has achieved better performance than classic machine learning in many research fields, as it employs more complex model architectures with stronger capability of extracting effective representations. Deep learning has been successfully applied to heart sound analysis in the past years. As most review works about heart sound analysis were given before 2017, the present survey is the first to work on a comprehensive overview to summarise papers on heart sound analysis with deep learning in the past six years 2017-2022. We introduce both classic machine learning and deep learning for comparison, and further offer insights about the advances and future research directions in deep learning for heart sound analysis.

Artificial intelligence for heart sound classification: A review

“…Some young, healthy individuals have S3 in their cardiac cycles, although S4 does not happen in a typical cardiac cycle . The presence of S3 may indicate heart failure (Ren et al, 2023), and the frequencies of S1 and S2 range from 20 to 200 Hz, while the frequencies of both S3 and S4 range from 15 to 65 Hz (Naseri & Homaeinezhad, 2013). Murmurs, which are additional sounds given by disturbed blood circulation, are significant in the diagnosis of cardiovascular disease.…”

“…In binary classification tasks, the more common evaluation indicators are sensitivity, specificity and accuracy (Ren et al, 2023). Precision and recall are frequently used in multi-class classification problems to assess how well a model performs for each class.…”

“…In addition, common heart sound denoising methods also include tunable-Q wavelet transform, adaptive noise canceller, discrete wavelet transform (DWT), variational mode decomposition, and combination of singular value decomposition (SVD) and compressed sensing, non-negative matrix factorization (NMF) with adaptive contour representation computation (ACRC) from the corresponding spectrogram. The study (Ren et al, 2023) states that to obtain better classification performance for a classifier that is not very powerful, segmentation may need to be used as an additional supplementary or as an internal process of a robust classifier. Experiments in Dissanayake et al (2020)…”

“…The DL is a subclass of ML that involves convolutional neural learning networks. In the past, several review studies on heart sound analysis have been carried out (Brites et al, 2021;Dwivedi et al, 2018;Ghosh, Nagarajan, & Tripathy, 2020;Kumar et al, 2022;Nabih-Ali et al, 2017), and the latest review study found in Google Scholar is Ren et al (2023). Dwivedi et al (2018) presented a systematic review and critical analysis of existing approaches for automated identification and classification of heart sounds.…”

“…Furthermore, this review summarizes traditional ML and DL approaches in denoising, segmentation, obtaining characteristics, and identifying cardiac sound signals. Compared to Ren et al (2023), we have added a discussion of DL denoising methods. We will also discuss issues requiring staff attention and future research directions in this survey, helping to advance research in heart sound analysis.…”