Heart Sound Signal Processing Techniques

Heart sound signal processing encompasses the acquisition, enhancement and analysis of phonocardiogram recordings to support non-invasive cardiac assessment. Signals are typically captured using electronic stethoscopes, yielding non-stationary waveforms composed of fundamental sounds (S1 and S2) and potential murmurs arising from turbulent blood flow. The principal stages in processing include denoising to suppress ambient and motion artefacts, segmentation to delineate individual cardiac cycles, feature extraction in time, frequency and time–frequency domains, and automated classification. Feature sets range from simple statistical descriptors to cepstral coefficients and wavelet-based decompositions. Classification techniques have evolved from support vector machines and k-nearest neighbours to deep convolutional and recurrent neural networks, often in end-to-end frameworks. Recent trends emphasise lightweight architectures for real-time analysis on mobile or point-of-care devices, explainable AI to increase clinical trust, and standardisation of datasets and evaluation metrics for reproducibility. These advances aim to broaden access to early screening for valvular diseases, congenital defects and general cardiac dysfunction, particularly in resource-limited settings and telemedicine applications.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent studies have undertaken a systematic comparison of shallow and deep learning models on a newly released heart sound corpus. By segmenting recordings into shorter auscultation-like intervals and introducing binary and multi-class classification tasks, this work benchmarked classical machine learning against state-of-the-art neural networks and analysed feature contributions to enhance interpretability. Another development describes a medical cyber–physical system for rapid on-site screening of valvular disorders. This integrated solution deploys several neural network models trained on a nine-category dataset, validated against open-access repositories, and optimised for mobile-device performance, achieving classification accuracies approaching clinical standards. Additionally, an international algorithmic challenge invited open-source methods for murmur and abnormal-function detection from phonocardiogram recordings. By sourcing thousands of paediatric recordings, defining cost-sensitive evaluation metrics, and mandating full code submission, the initiative highlighted diverse traditional and deep learning approaches, advanced reproducibility and underscored the promise of algorithmic screening in underserved regions.

Heart Sound Signal Processing Techniques publication trend

The graph below shows the total number of articles in heart sound signal processing techniques across all publications each year (not limited to Nature Index journals).

Technical terms

Phonocardiogram (PCG): Digital recording of heart sounds acquired via an electronic stethoscope for computational analysis.

Denoising: Signal-processing step to reduce ambient, respiratory and motion artefacts in recorded heart sounds.

Segmentation: Division of a continuous PCG signal into individual cardiac cycles or heart sound components (S1, systole, S2, diastole).

Feature extraction: Transformation of preprocessed signals into representative numerical descriptors, such as time-domain statistics, spectral coefficients or wavelet features.

Convolutional neural network (CNN): Deep learning architecture that applies learnable filters to extract hierarchical features from input signals.

Recurrent neural network (RNN): Neural architecture designed to capture temporal dependencies in sequential data such as heart sound waveforms.

Cyber–Physical System (CPS): Integrated computational and physical system enabling real-time data acquisition, processing and feedback in medical devices.

References

1. Learning Representations from Heart Sound: A Comparative Study on Shallow and Deep Models. Cyborg and Bionic Systems (2024).
2. An intelligent Medical Cyber–Physical System to support heart valve disease screening and diagnosis. Expert Systems with Applications (2024).
3. Heart murmur detection from phonocardiogram recordings: The George B. Moody PhysioNet Challenge 2022. PLOS Digital Health (2023).