Separation of heart sound signal from noise in joint cycle frequency-time-frequency domains based on fuzzy detection

An edge-device-compatible algorithm for valvular heart diseases screening using phonocardiogram signals with a lightweight convolutional neural network and self-supervised learning

BACKGROUND AND OBJECTIVES

Detection and classification of heart murmur using mobile-phone-collected sound is an emerging approach to the scale-up screening of valvular heart disease at a population level. Nonetheless, the widespread adoption of artificial intelligence (AI) methods for this type of mobile health (mHealth) application requires highly accurate and lightweight AI models that can be deployed in consumer-grade mobile devices. This study presents a lightweight deep learning model and a self-supervised learning (SSL) method to utilise unlabelled data to improve the accuracy of valvular heart disease classification using phonocardiogram data.

METHODS

This study proposes a lightweight convolutional neural network (CNN) that consists of ten times fewer parameters than other deep learning models to classify phonocardiogram data. SSL is applied to harness a large collection of unlabelled data as pre-training to enhance the accuracy and robustness of the model and reduce the number of epochs required to converge. A mobile application prototype that encapsulates the model is developed to perform in-device inference and fine-turning.

RESULTS

The proposed lightweight model achieves an average accuracy of 98.65% in 10-fold cross-validation. When coupled with SSL using unlabelled data, the pre-trained model can reach an average accuracy higher than 99.4% in 10-fold cross-validation. Furthermore, SSL-trained models have a 4-20% improvement in classification accuracy over non-SSL-trained models when tested with perturbed or noisy data, suggesting that SSL improves robustness of the model. When deployed on common smartphones, in-device fine-tuning and inference of the model can be completed within 0.03-0.37 s, which is considerably faster than 0.22-5.7 s by a standard CNN model that have ten times the number of parameters. Our lightweight model also consumes only a third of the power compared to the larger standard model.

CONCLUSION

This work presents a lightweight and accurate phonocardiogram classifier that supports near real-time performance on standard mobile devices.

Monaural cardiopulmonary sound separation via complex-valued deep autoencoder and cyclostationarity

Objective.Cardiopulmonary auscultation is promising to get smart due to the emerging of electronic stethoscopes. Cardiac and lung sounds often appear mixed at both time and frequency domain, hence deteriorating the auscultation quality and the further diagnosis performance. The conventional cardiopulmonary sound separation methods may be challenged by the diversity in cardiac/lung sounds. In this study, the data-driven feature learning advantage of deep autoencoder and the common quasi-cyclostationarity characteristic are exploited for monaural separation.Approach.Different from most of the existing separation methods that only handle the amplitude of short-time Fourier transform (STFT) spectrum, a complex-valued U-net (CUnet) with deep autoencoder structure, is built to fully exploit both the amplitude and phase information. As a common characteristic of cardiopulmonary sounds, quasi-cyclostationarity of cardiac sound is involved in the loss function for training.Main results. In experiments to separate cardiac/lung sounds for heart valve disorder auscultation, the averaged achieved signal distortion ratio (SDR), signal interference ratio (SIR), and signal artifact ratio (SAR) in cardiac sounds are 7.84 dB, 21.72 dB, and 8.06 dB, respectively. The detection accuracy of aortic stenosis can be raised from 92.21% to 97.90%.Significance. The proposed method can promote the cardiopulmonary sound separation performance, and may improve the detection accuracy for cardiopulmonary diseases.

Design of Abnormal Heart Sound Recognition System Based on HSMM and Deep Neural Network

Introduction

Heart sound signal is an important physiological signal of human body, and the identification and research of heart sound signal is of great significance.

Methods

For abnormal heart sound signal recognition, an abnormal heart sound recognition system, combining hidden semi-Markov models (HSMM) with deep neural networks, is proposed. Firstly, HSMM is used to build a heart sound segmentation model to accurately segment the heart sound signal, and then the segmented heart sound signal is subjected to feature extraction. Finally, the trained deep neural network model is used for recognition.

Results

Compared with other methods, this method has a relatively small amount of input feature data and high accuracy, fast recognition speed.

Discussion

HSMM combined with deep neural network is expected to be deployed on smart mobile devices for telemedicine detection.

Automated Signal Quality Assessment for Heart Sound Signal by Novel Features and Evaluation in Open Public Datasets

Automated heart sound signal quality assessment is a necessary step for reliable analysis of heart sound signal. An unavoidable processing step for this objective is the heart sound segmentation, which is still a challenging task from a technical viewpoint. In this study, ten features are defined to evaluate the quality of heart sound signal without segmentation. The ten features come from kurtosis, energy ratio, frequency-smoothed envelope, and degree of sound periodicity, where five of them are novel in signal quality assessment. We have collected a total of 7893 recordings from open public heart sound databases and performed manual annotation for each recording as gold standard quality label. The signal quality is classified based on two schemes: binary classification (“unacceptable” and “acceptable”) and triple classification (“unacceptable”, “good,” and “excellent”). Sequential forward feature selection shows that the feature “the degree of periodicity” gives an accuracy rate of 73.1% in binary SVM classification. The top five features dominate the classification performance and give an accuracy rate of 92%. The binary classifier has excellent generalization ability since the accuracy rate reaches to (90.4 ± 0.5) % even if 10% of the data is used to train the classifier. The rate increases to (94.3 ± 0.7) % in 10-fold validation. The triple classification has an accuracy rate of (85.7 ± 0.6) % in 10-fold validation. The results verify the effectiveness of the signal quality assessment, which could serve as a potential candidate as a preprocessing in future automatic heart sound analysis in clinical application.

A Markov-Switching Model Approach to Heart Sound Segmentation and Classification

OBJECTIVE

We consider challenges in accurate segmentation of heart sound signals recorded under noisy clinical environments for subsequent classification of pathological events. Existing state-of-the-art solutions to heart sound segmentation use probabilistic models such as hidden Markov models (HMMs), which, however, are limited by its observation independence assumption and rely on pre-extraction of noise-robust features.

METHODS

We propose a Markov-switching autoregressive (MSAR) process to model the raw heart sound signals directly, which allows efficient segmentation of the cyclical heart sound states according to the distinct dependence structure in each state. To enhance robustness, we extend the MSAR model to a switching linear dynamic system (SLDS) that jointly model both the switching AR dynamics of underlying heart sound signals and the noise effects. We introduce a novel algorithm via fusion of switching Kalman filter and the duration-dependent Viterbi algorithm, which incorporates the duration of heart sound states to improve state decoding.

RESULTS

Evaluated on Physionet/CinC Challenge 2016 dataset, the proposed MSAR-SLDS approach significantly outperforms the hidden semi-Markov model (HSMM) in heart sound segmentation based on raw signals and comparable to a feature-based HSMM. The segmented labels were then used to train Gaussian-mixture HMM classifier for identification of abnormal beats, achieving high average precision of 86.1% on the same dataset including very noisy recordings.

CONCLUSION

The proposed approach shows noticeable performance in heart sound segmentation and classification on a large noisy dataset.

SIGNIFICANCE

It is potentially useful in developing automated heart monitoring systems for pre-screening of heart pathologies.

Design and evaluation of a parametric model for cardiac sounds

Heart sound analysis plays an important role in the auscultative diagnosis process to detect the presence of cardiovascular diseases. In this paper we propose a novel parametric heart sound model that accurately represents normal and pathological cardiac audio signals, also known as phonocardiograms (PCG). The proposed model considers that the PCG signal is formed by the sum of two parts: one of them is deterministic and the other one is stochastic. The first part contains most of the acoustic energy. This part is modeled by the Matching Pursuit (MP) algorithm, which performs an analysis-synthesis procedure to represent the PCG signal as a linear combination of elementary waveforms. The second part, also called residual, is obtained after subtracting the deterministic signal from the original heart sound recording and can be accurately represented as an autoregressive process using the Linear Predictive Coding (LPC) technique. We evaluate the proposed heart sound model by performing subjective and objective tests using signals corresponding to different pathological cardiac sounds. The results of the objective evaluation show an average Percentage of Root-Mean-Square Difference of approximately 5% between the original heart sound and the reconstructed signal. For the subjective test we conducted a formal methodology for perceptual evaluation of audio quality with the assistance of medical experts. Statistical results of the subjective evaluation show that our model provides a highly accurate approximation of real heart sound signals. We are not aware of any previous heart sound model rigorously evaluated as our proposal.

A Noise Reduction Technique Based on Nonlinear Kernel Function for Heart Sound Analysis

The main difficulty encountered in interpretation of cardiac sound is interference of noise. The contaminated noise obscures the relevant information, which are useful for recognition of heart diseases. The unwanted signals are produced mainly by lungs and surrounding environment. In this paper, a novel heart sound denoising technique has been introduced based on a combined framework of wavelet packet transform and singular value decomposition (SVD). The most informative node of the wavelet tree is selected on the criteria of mutual information measurement. Next, the coefficient corresponding to the selected node is processed by the SVD technique to suppress noisy component from heart sound signal. To justify the efficacy of the proposed technique, several experiments have been conducted with heart sound dataset, including normal and pathological cases at different signal to noise ratios. The significance of the method is validated by statistical analysis of the results. The biological information preserved in denoised heart sound signal is evaluated by the k-means clustering algorithm. The overall results show that the proposed method is superior than the baseline methods.

An open access database for the evaluation of heart sound algorithms

In the past few decades, analysis of heart sound signals (i.e. the phonocardiogram or PCG), especially for automated heart sound segmentation and classification, has been widely studied and has been reported to have the potential value to detect pathology accurately in clinical applications. However, comparative analyses of algorithms in the literature have been hindered by the lack of high-quality, rigorously validated, and standardized open databases of heart sound recordings. This paper describes a public heart sound database, assembled for an international competition, the PhysioNet/Computing in Cardiology (CinC) Challenge 2016. The archive comprises nine different heart sound databases sourced from multiple research groups around the world. It includes 2435 heart sound recordings in total collected from 1297 healthy subjects and patients with a variety of conditions, including heart valve disease and coronary artery disease. The recordings were collected from a variety of clinical or nonclinical (such as in-home visits) environments and equipment. The length of recording varied from several seconds to several minutes. This article reports detailed information about the subjects/patients including demographics (number, age, gender), recordings (number, location, state and time length), associated synchronously recorded signals, sampling frequency and sensor type used. We also provide a brief summary of the commonly used heart sound segmentation and classification methods, including open source code provided concurrently for the Challenge. A description of the PhysioNet/CinC Challenge 2016, including the main aims, the training and test sets, the hand corrected annotations for different heart sound states, the scoring mechanism, and associated open source code are provided. In addition, several potential benefits from the public heart sound database are discussed.

Phonocardiogram signal compression using sound repetition and vector quantization

BACKGROUND

A phonocardiogram (PCG) signal can be recorded for long-term heart monitoring. A huge amount of data is produced if the time of a recording is as long as days or weeks. It is necessary to compress the PCG signal to reduce storage space in a record and play system. In another situation, the PCG signal is transmitted to a remote health care center for automatic analysis in telemedicine. Compression of the PCG signal in that situation is necessary as a means for reducing the amount of data to be transmitted. Since heart beats are of a cyclical nature, compression can make use of the similarities in adjacent cycles by eliminating repetitive elements as redundant. This study proposes a new compression method that takes advantage of these repetitions.

METHODS

Data compression proceeds in two stages, a training stage followed by the compression as such. In the training stage, a section of the PCG signal is selected and its sounds and murmurs (if any) decomposed into time-frequency components. Basic components are extracted from these by clustering and collected to form a dictionary that allows the generative reconstruction and retrieval of any heart sound or murmur. In the compression stage, the heart sounds and murmurs are reconstructed from the basic components stored in the dictionary. Compression is made possible because only the times of occurrence and the dictionary indices of the basic components need to be stored, which greatly reduces the number of bits required to represent heart sounds and murmurs. The residual that cannot be reconstructed in this manner appears as a random sequence and is further compressed by vector quantization. What we propose are quick search parameters for this vector quantization.

RESULTS

For normal PCG signals the compression ratio ranges from 20 to 149, for signals with median murmurs it ranges from 14 to 35, and for those with heavy murmurs, from 8 to 20, subject to a degree of distortion of ~5% (in percent root-mean-square difference) and a sampling frequency of 4kHz.

DISCUSSION

We discuss the selection of the training signal and the contribution of vector quantization. Performance comparisons between the method proposed in this study and existing methods are conducted by computer simulations.

CONCLUSIONS

When recording and compressing cyclical sounds, any repetitive components can be removed as redundant. The redundancies in the residual can be reduced by vector quantization. The method proposed in this study achieves a better performance than existing methods.

Pediatric heart sound segmentation using hidden Markov model

Recent advances in technology have enabled automatic cardiac auscultation using digital stethoscopes. This in turn creates the need for development of algorithms capable of automatic segmentation of heart sounds. Pediatric heart sound segmentation is a challenging task due to various confounding factors including the significant influence of respiration on children's heart sounds. The current work investigates the application of homomorphic filtering and Hidden Markov Model for the purpose of segmenting pediatric heart sounds. The efficacy of the proposed method is evaluated on the publicly available Pascal Challenge dataset and its performance is compared with those of three other existing methods. The results show that our proposed method achieves an accuracy of 92.4%±1.1% and 93.5%±1.1% in identifying the first and second heart sound components, respectively, and is superior to three other existing methods in terms of accuracy or computational complexity.

USING SHORT-TIME FOURIER TRANSFORM AND WAVELET PACKET TRANSFORM TO ATTENUATE NOISE FROM HEART SOUND SIGNAL FOR WEARABLE E-HEALTHCARE DEVICE

A soft-computing method attenuating noise from heart sound (HS) signal for wearable e-healthcare device is proposed. The HS signal is decomposed by third-level wavelet packet transform (WPT). An automatic HS cycle detection algorithm is applied to find HS cycles in the (3, 0) leaf signal of WPT tree. Based on the quasi-cyclic feature of HS, short-time Fourier transform is implemented for cycles of each WPT tree leaf signal to decompose each cycle into time-frequency fragments which are called particles. Furthermore, the novel cuboid method is proposed to identify constituents of HS and noise from such generated particles. The particles representing HS are then retained and merged into noise-quasi-free WPT tree leaf signals. Eventually the inverse WPT (IWPT) is employed to build the noise-quasi-free HS signal. The method is assessed using mean square error (MSE) and compared with wavelet multi-threshold method (WMTM) and Tang's method. The experimental results show that the proposed method not only filters HS signal effectively but also well retains its pathological information.

A robust method for online heart sound localization in respiratory sound based on temporal fuzzy c-means

This work presents a detailed framework to detect the location of heart sound within the respiratory sound based on temporal fuzzy c-means (TFCM) algorithm. In the proposed method, respiratory sound is first divided into frames and for each frame, the logarithmic energy features are calculated. Then, these features are used to classify the respiratory sound as heart sound (HS containing lung sound) and non-HS (only lung sound) by the TFCM algorithm. The TFCM is the modified version fuzzy c-means (FCM) algorithm. While the FCM algorithm uses only the local information about the current frame, the TFCM algorithm uses the temporal information from both the current and the neighboring frames in decision making. To measure the detection performance of the proposed method, several experiments have been conducted on a database of 24 healthy subjects. The experimental results show that the average false-negative rate values are 0.8 ± 1.1 and 1.5 ± 1.4 %, and the normalized area under detection error curves are 0.0145 and 0.0269 for the TFCM method in the low and medium respiratory flow rates, respectively. These average values are significantly lower than those obtained by FCM algorithm and by the other compared methods in the literature, which demonstrates the efficiency of the proposed TFCM algorithm. On the other hand, the average elapsed time of the TFCM for a data with length of 0.2 ± 0.05 s is 0.2 ± 0.05 s, which is slightly higher than that of the FCM and lower than those of the other compared methods.

Heart monitoring systems--a review

To diagnose health status of the heart, heart monitoring systems use heart signals produced during each cardiac cycle. Many types of signals are acquired to analyze heart functionality and hence several heart monitoring systems such as phonocardiography, electrocardiography, photoplethysmography and seismocardiography are used in practice. Recently, focus on the at-home monitoring of the heart is increasing for long term monitoring, which minimizes risks associated with the patients diagnosed with cardiovascular diseases. It leads to increasing research interest in portable systems having features such as signal transmission capability, unobtrusiveness, and low power consumption. In this paper we intend to provide a detailed review of recent advancements of such heart monitoring systems. We introduce the heart monitoring system in five modules: (1) body sensors, (2) signal conditioning, (3) analog to digital converter (ADC) and compression, (4) wireless transmission, and (5) analysis and classification. In each module, we provide a brief introduction about the function of the module, recent developments, and their limitation and challenges.

Modeling of heart sound morphology and analysis of the morphological variations induced by respiration

In this study, each peak/valley of a heart sound was modeled by a Gaussian curve and characterized by amplitude, timing, and supporting width. This model was applied to analyze the morphological variations induced by respiration in 12 subjects. It was observed that the morphology exhibited regular behaviors with respiration. The amplitude of the prominent peaks and valleys of S2 (the second heart sound) were commonly attenuated during expiration and were accentuated during inspiration whereas no consistent observations were obtained for S1 (the first heart sound). The supporting width of S1 commonly decreased with expiration and increased with inspiration whereas the supporting width of S2 displayed no significant changes during respiration. For all subjects, the delay of S1 increased during inspiration and decreased during expiration. However, the delay of the aortic component increased during expiration and decreased during inspiration. The pulmonary component of S2 was observed in 7 of 12 subjects, and the delay was opposite to that of the aortic component. The opposing delays yielded a splitting between the two components of S2 that increased during inspiration and decreased during expiration. The delay pattern was the most consistent observation in all subjects. These results suggest that a quantitative analysis of morphological variations, particularly the delay pattern, could be used as a non-invasive continuous monitoring method of hemodynamic change during respiratory cycles.

Optimum heart sound signal selection based on the cyclostationary property

Noise often appears in parts of heart sound recordings, which may be much longer than those necessary for subsequent automated analysis. Thus, human intervention is needed to select the heart sound signal with the best quality or the least noise. This paper presents an automatic scheme for optimum sequence selection to avoid such human intervention. A quality index, which is based on finding that sequences with less random noise contamination have a greater degree of periodicity, is defined on the basis of the cyclostationary property of heart beat events. The quality score indicates the overall quality of a sequence. No manual intervention is needed in the process of subsequence selection, thereby making this scheme useful in automatic analysis of heart sound signals.