Search

A multi-channel stethograph system was designed and developed as an electronic auscultation system for graphic recording of heart, lung and trachea (HLT) sounds non-invasively through a set of 16 acoustic sensors. The multi-channel stethograph system was fabricated by placing 16 microphone based acoustic sensor in a CNC machined Delrin® housing cases that are covered using diaphragms. Among the 16 acoustic sensors, 14 were positioned in a memory foam pad and 2 were placed directly on the heart and trachea to acquire sounds simultaneously from the lungs, heart and trachea. The sounds acquired from the 16 acoustic sensors were processed through a custom designed and fabricated 16-channel PCB for signal conditioning. A National Instruments (NI) 9205 data acquisition device (DAQ) along with a NI 9191 wireless chassis was used to acquire and wirelessly transmit the data from the 16-channel PCB to a Wi-Fi enabled device such as a PC/tablet. A custom LabVIEW program was developed on a Wi-Fi enabled PC/tablet to record the data from the DAQ. In addition, a MATLAB program was developed to convert the recorded data from the acoustic sensors into 16 audio files (for audio playback) and plot the waveforms in time and frequency domain as well as spectrogram for visual examination of any abnormal patterns in inhalation and exhalation. This provides critical information on the presence of wheezes, crackles and rhonchi sounds as well as abnormal heartbeat and respiration rate which helps in analyzing the condition of heart and lungs. The graphically displayed HLT sounds will help physicians in the clinical diagnosis and monitoring of lung and heart disorders, particularly chronic obstructive pulmonary disease (COPD), asthma, pneumonia, and congestive heart failure by providing objective evidence.

Heartbeats are important aspects for the study of heart diseases in medical sciences as they provide vital information on heart disorders and abnormalities in heart rhythms. Each heartbeat provides a QRS complex in the electrocardiogram (ECG) which is centered at the R-peak. The analysis of ECG is hindered by low-frequency noise, high-frequency noise, interference from P and T waves, and changes in QRS morphology. This paper presents a new peak detection algorithm that can suppress the noise and adapt to changes in ECG signal morphology for better detection performance. The proposed algorithm is based on median and moving average (MA) filtering, segmentation, time and amplitude thresholds, and statistical false peak elimination (SFPE). The filters are first used in preprocessing to reduce unwanted noise and interference. The data is then divided into smaller segments and each segment is then analyzed using two distinct thresholds, a time axis (x-axis) threshold and an amplitude (y-axis) threshold. Next, the false peaks are eliminated resulting from any residue of noise using an average value of peak-to-peak interval. A post-processing stage is added to eliminate any peak that is detected twice and to search for missed low-amplitude peaks. The proposed method is tested on MIT-BIH arrhythmia and Fantasia databases and provides better results in comparison to several state-of-the-art methods in the field. The mean sensitivity, positive predictivity, and detection error rates for the proposed method are 99.82%, 99.88%, and 0.31%, respectively, for the MIT-BIH arrhythmia database and 99.92%, 99.90%, and 0.18%, respectively, for the Fantasia database.

Heart arrhythmia have been a life-threatening disease to human for a very long time, many techniques and methods have been developed by human experts since the advent of ECG machine, these approaches work well and most of the time patients with heart diseases can be diagnosed accurately, and yet not timely, human experts for heart disease are far from adequate. Deep learning models for heartbeat arrhythmia detection, therefore, have been proposed and investigated during the past few years, they have good results on some data-sets concerning arrhythmia detection and classification, some even outperformed human doctors. However, like most deep learning models, the underlying mechanism has not been justified and explained very well, for example, we cannot interpret the decision momentum of certain models. Concerning that, we developed a general feature extraction framework for ECG data, it can perform various kinds of feature engineering tasks, all of the features have their meaning under a clinical context. We distilled the features on a large ECG data-set, then these features were fed into a Bayesian Neural Network for training. The BNN has a sparsity-inducing prior distribution arranged in a tied manner to learn the importance of every feature concerning the outcome. We evaluated this method for a classification problem and made a comparison between the way our model makes decisions and how field experts do for the same problem, we were able to find out that a) the features extracted from raw ECG data using our framework can be used as an indication for specific heartbeat arrhythmia, b) the mechanism of the model can be interpreted as a decision making procedure by weighing relative metrics just like human experts do, c) the weight of the features extracted from raw ECG data can be used to build a knowledge tree for guidance on diagnosing of certain heartbeat disease.

In this study, a new approach using a novel ensemble classification algorithm based on ECG morphological features is proposed for accurate detection of heart ventricular and atrial abnormalities. First, the raw ECG signal is preprocessed and the main character waves are detected. Second, a combination of ECG morphological features is proposed and extracted from the selected ECG segments. The proposed feature set contains morphological parameters, morphological visual pattern of QRS complex, and principle components of the third level and fourth level of a four-level Sym8 wavelet-decomposed ECG waveform. Next, a novel ensemble classification algorithm, with the key idea of integrating the knowledge acquired by several popular classification algorithms for this task into an ensemble system, is proposed so that the accuracy and robustness over various arrhythmia types could be improved. Finally, the features are applied to the proposed ensemble classification algorithm for abnormality detection. The proposed approach achieved an overall accuracy of 98.68% when it was validated on fifteen heartbeat types from the MIT-BIH arrhythmia database (MITDB), according to the Association for Advancement of Medical Instrumentation (AAMI) standard. The classification accuracies of the six main types – normal beat (N), right bundled branch blocks beat (R), left bundled branch blocks beat (L), atrial premature beat (A), premature ventricular contractions beat (V), and paced beat (P) are 98.75%, 99.77%, 99.70%, 94.81%, 98.57%, and 99.94%, respectively. The proposed approach proves a solid result in comparison with component classification algorithms as well as recent peer works.

Daily vital signs monitoring is very important for detecting diseases in early stages and for preventive treatments. Such a task can be achieved by taking advantage of the omnipresence of cameras in people's personal space. As heart-related diseases are part of the leading causes of deaths worldwide, monitoring heart-related vital signs appear to be very crucial. In this article we aim to provide a touchless approach and propose a robust method for estimating heart rate through analysis of face videos. In particular, we consider a challenging scenario, i.e., the user is on a video call and may often move his/her head. Existing touchless, vision-based methods use either photoplethysmography (PPG) or ballistocardiography (BCG). PPG methods exploit color changes in human skin during heartbeats caused by blood volume variations, but this is very sensitive to unstable lighting conditions. On the other hand, BCG methods exploit subtle head motions caused by Newtonian reaction to blood influx into the head at each heartbeat, thus being sensitive to a subject's voluntary head movements. Unlike conventional studies where either a PPG method or a BCG method is used, we propose to combine both to overcome the weakness faced by each method. We use BCG methods as the main approach due to their better accuracy on heart rate estimation, and PPG methods are used as the secondary backup to improve the accuracy in cases of large and frequent voluntary head movements. To this end, we introduce a dynamic voting system that effectively combines results of several variants of PPG and BCG methods. Experiments conducted on 20 healthy subjects with different skin tones in different lighting conditions show that our method has better accuracy compared to state-of-the-art methods, well addressing large voluntary head movements. Our method had a mean absolute error of 1.23 beats per minute (BPM) in the cases without voluntary head movements and 2.78 BPM in the cases with voluntary head movements Refereed/Peer-reviewed

Remote techniques for measuring human vital signs have attracted great interests due to the benefits shown in medical monitoring and military applications. Compared with continuous-wave Doppler radar, frequency-modulated continuous-wave (FMCW) radar which can discriminate vital signs from different distances, shows potential for reducing the interferences from other targets and the environment. However, in the state-of-the-art algorithms, only one chirp per frame is utilized for FMCW-based vital sign monitoring. Moreover, the vital signal is extracted from only one range bin of the fast Fourier transform. Which does not make full utilization of the long system idle time, and loses the power distributed on other range bins. By exploiting the relationship between respiration and heartbeat vibrations, an adaptive identification embedded ensemble empirical mode decomposition (EEMD) method for joint-range spectral estimation is proposed to measure the heart rate. First, A multi-chirp processing is presented for a 2-dimensional phase accumulation. Then the phase signals from a sequence of range bins are decomposed with a fast adaptive identification process. Finally, with the identified heartbeat components, we solve a multiple measurement vectors problem to estimate the heart rate. Experimental results showed that, at the detection range from 1 m~ 2.5 m, the proposed method can robustly distinguish the heartbeat from respiration and its harmonics and accurately estimate the heart rate with a root mean square error less than 6 bpm.

Heart rate variability (HRV) is a widely used measure for the variation of the heartbeat intervals controlled by the autonomic nervous system (ANS), which is primarily obtained by electrocardiogram (ECG) signals. In general, HRV of a healthy person exhibits long-range correlations with dynamic fluctuations, whether the complexity of HRV decreases with aging and incidence of disease. Recently, the complexity of differential inter-beat intervals, referred to differential R-R intervals, is known to be more effective than original R-R intervals to reflect HRV. The multiscale based entropy methods have been developed to quantify HRV using R-R intervals. In spite of their capability, it still remains unreliable quantification of HRV. Here, we propose a new multiscale complexity quantification measure with differential R-R intervals for HRV analysis. To verify the performance of the proposed method, we evaluate the complexity of differential HRV extracted from ECG signals of congestive heart failure (CHF) patients and healthy subjects. The results show that multiscale distribution entropy (MDE) of differential R-R interval has improved capability for quantifying the complexity of HRV regardless of the length of time series.

Motion artifacts may occur in coronary computed tomography angiography (CCTA) due to the heartbeat and impede the clinician's diagnosis of coronary arterial diseases. Thus, motion artifact correction of the coronary artery is required to quantify the risk of disease more accurately. We present a novel method based on deep learning for motion artifact correction in CCTA. Because the image of the coronary artery without motion (the ground-truth data required in supervised deep learning) is medically unattainable, we apply a style transfer method to 2D image patches cropped from full-phase 4D computed tomography (CT) to synthesize these images. We then train a convolutional neural network (CNN) for motion artifact correction using this synthetic ground-truth (SynGT). During testing, the output motion-corrected 2D image patches of the trained network are reinserted into the 3D CT volume with volumetric interpolation. The proposed method is evaluated using both phantom and clinical data. A phantom study demonstrates comparable results to other methods in quantitative performance and outperforms those methods in computation time. For clinical data, a quantitative analysis based on metric measurements is presented that confirms the correction of motion artifacts. Moreover, an observer study finds that by applying the proposed method, motion artifacts are markedly reduced, and boundaries of the coronary artery are much sharper, with a strong inter-observer agreement (κ = 0.78). Finally, evaluations using commercial software on the original and resulting CT volumes of the proposed method reveal a considerable increase in tracked coronary artery length.

An Electrocardiogram (ECG) is a typical method used to detect heartbeat, and an ECG signal analysis enables the detection of some heart diseases. However, the ECG-based heartbeat detection requires device attachment, which is not preferred for daily use. A Doppler sensor could be a device used to enable the non-contact heartbeat detection. In this paper, we propose a Doppler sensor-based ECG signal reconstruction method by a hybrid deep learning model with Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). An ECG signal can be reconstructed by relating features of a heartbeat signal obtained by a Doppler sensor to those of the ECG signal. Thus, we construct the deep learning model that extracts the spatial and temporal features from the heartbeat signal by CNN and LSTM. Based on the extracted features, the ECG signal is reconstructed. We conducted experiments to observe heartbeat against 9 healthy subjects without heart disease. The experimental results showed that our method performed ECG signal reconstruction with a correlation coefficient of 0.86 between the reconstructed and actual ECG signals, even without attaching devices. The results indicate that it is possible to remotely reconstruct an ECG signal from a heartbeat signal via a Doppler sensor.

Systolic and diastolic blood pressure estimation using arm-cuff monitors is one of the most common cardiovascular evaluation criteria in healthcare today, however these measures lack critical heartbeat and pressure dynamics. Pulse-transit time can be used as an alternative for arm-cuff monitors, but gives only one parameter for two pressure quantities. Ultrasound, computed tomography scan, and magnetic resonance imaging can retrieve geometrical features of the heart, but cannot directly estimate vascular pressures. Here we show a novel radio-frequency heartbeat sensor based on multi-point near-field observation. By comparing to synchronized electrocardiogram and auscultation, the multi-point sensor can assess motion and pressure in different parts of the heart following the Wiggers diagram. By applying the Hilbert-Huang frequency-time transform, the central blood pressure can be derived from the vascular vibration characteristics as continuous transients, including during the pulmonary cycle which is previously inaccessible from branchial measurements. Our scheme can be further extended to a full multiple-input-multiple-output (MIMO) channel arrangement, producing rich content for diagnostic and biometric applications. We employ dynamic time warping analyses to illustrate the sensor's wealth of diversified information across different channels and persons. This new multi-point sensor, which can be worn conveniently over clothing, enables unprecedented monitoring capabilities of detailed central blood pressure transients and heartbeat dynamics under the given measurement instructions for at-home and clinical cardiovascular diagnostics.

The feasibility of monitoring driving psychological fatigue through unconstrained heartbeat extraction by using pressure sensors array was investigated. The pressure signals were obtained by close contact between subject's back and backrest of car seat. Correlation coefficients of electrocardiogram signals and pressure signals were used to select the optimal monitoring point due to the unavoidable adjustment of sitting posture during long times of driving. Lifting wavelet transform was applied for de-noising. J-J interval signals, which represent heartbeat signals, were extracted. The correlation dimension of heartbeat signals, reflecting the complicated degree and the chaotic level of a nonlinear dynamic system, was used as the indicator to determine the psychological fatigue. Its validity was verified by the consistency of correlation dimension with R-R interval (represents cardiac cycle). Finally, questionnaires during driving were administered. The effectiveness of this method to judge driver psychological fatigue was confirmed. This finding indicates that driving psychological fatigue can be monitored through unconstrained heartbeat signal extraction by using the pressure sensor array.

Extreme learning machine based on local receptive fields (ELM-LRFs) is a very fast method that can be used for feature extraction and classification. Bidirectional long-short time memory network (BLSTM), a widely used type of recurrent neural network (RNN) architecture, has showed excellent performance in time series processing fields. In this paper, we combine the superiority of above algorithms and propose a fast and accurate hybrid deep learning model which is named DELM-LRF-BLSTM for ECG signal recognition. This model uses the segmented heartbeats as input and employs a deep ELM-LRF to gain significant local spatial features. Then we fed the features to a three-layer BLSTM and it can extract temporal features for ECG signal recognition. The combination of ELM-LRF and BLSTM can not only consider the local information in a heartbeat, but also consider the long-distance dependence between heartbeats. Experimental results on MIT-BIH Arrhythmia dataset show that the proposed DELM-LRF-BLSTM algorithm has high accuracy and sensitivity, up to 99.32% and 97.15% respectively, which verifies the effectiveness and feasibility of the model. Moreover, only 6.1 millisecond is needed for once heartbeat recognition operation. Due to it's high performance and low computational complexity, the proposed algorithm is feasible for practical use.

This paper aims at the analysis of the VdP heartbeat mathematical model. We have analysed the conditionality of a mathematical model which represents the oscillatory behaviour of the heart. A novel neuroevolutionary approach is chosen to analyse the mathematical model. The characteristics of the cardiac pulse of the heart are examined by considering two major scenarios with sixteen different cases. Artificial neural networks (ANNs) are constructed to obtain the best solutions for the heartbeat model. Unknown weights are finely tuned by a combination of a global search technique the Harris Hawks Optimizer (HHO) and a local search technique the Interior Point Algorithm (IPA). Stable behaviour of solutions obtained by considering different cases demonstrates that the model under consideration is well-conditioned. The accuracy of our novel procedure is established by getting the lowest residual errors in our solution for all cases. Graphical and statistical analysis are added to further elaborate the accuracy of our approach.

Heartbeat detection is one of key techniques to monitor our health condition in daily life, and demands for this technique have increased year and year. Thanks to the non-contact and non-invasive features, various Doppler sensor-based detection methods have been investigated so far. However, the heartbeat detection accuracy of the conventional methods could get degraded due to the low SNR (Signal-to-Noise Ratio) of heartbeat components. Thus, even after some signal processing, non-heartbeat components still remain over such processed signal, which could degrade the heartbeat detection accuracy. In particular for the subjects with low HR (Heart Rate), the estimated HR tends to be higher than the ground truth HR due to such non-heartbeat components, though the conventional methods have mainly focused on the heartbeat detection against the subjects with the normal HR higher than 50 bpm (Beats Per Minute). In this paper, to accurately detect heartbeat even with low HR via a Doppler sensor, we propose a heartbeat detection method based on heartbeat signal reconstruction with convolutional LSTM (Bidirectional-Long Short-Term Memory). In the proposed method, to reconstruct a heartbeat signal based on the periodicity of heartbeat and the spectrum distribution peculiar to heartbeat, successive spectrograms that might be due to heartbeat is used as an input to convolutional LSTM. In addition, for better reconstruction of a heartbeat signal, the previously estimated RRI (R-R Interval) is also used as a feature in the proposed deep learning model with convolutional LSTM. Through the experiments, we confirmed that our proposed method accurately detected heartbeat against 17 subjects including the ones with the HR lower than 50 bpm.

Heart rate variability (HRV) can be used as a common detection method for congestive heart failure (CHF). Existing researches regarding HRV, including both linear indicators and nonlinear characteristics, are mostly based on the $RR$ intervals of the ECG signal. This article proposed a sequence that can reflect the regulation of sympathetic and parasympathetic nerve on heart rate, and on this basis, conducted multifractal detrended fluctuation analysis (MFDFA). We extracted multifractal features to quantitatively compare the complexity of proposed sequence between the healthy and CHF groups. Results showed that abnormal physiological and pathological conditions due to the weakening of autonomic nerve control can reduce the complexity of the heartbeat signal. Estimate the separation performance of all features, the best discrimination is obtained for the area under the mass index spectrum $S1_\tau $ as providing 100% accuracy in separating the Healthy Young and CHF groups, and 90.93% separation accuracy between the Healthy Elderly and CHF groups. This work provide a good basis for the diagnosis of CHF with a novel perspective.

Electrocardiogram (ECG) data recorded by Holter monitors are extremely hard to analyze manually. Therefore, it is necessary to automatically analyze and categorize each heartbeat using a computer-aid method. Because convolutional neural networks (CNNs) can classify ECG signals automatically without trivial manual feature extractions, they have received extensive attention. However, it is anticipated that improving the existing CNN classifiers might provide better overall accuracy, sensitivity, positive predictivity, etc. In this study, we proposed a CNN based ECG heartbeat classification method. Based on the MIT-BIH arrhythmia database, our proposed method achieved a sensitivity of 99.2% and positive predictivity of 99.4% in VEB detection; a sensitivity of 97.5% and positive predictivity of 99.1% in SVEB detection; and an overall accuracy of 99.43%. Our proposed system can be directly implemented on wearable devices to monitor long-term ECG data.

This paper presents an important advancement in heart activity monitoring, focusing on non-contact sensor data, which tend to be noisy due to interference, and the limitations of non-contact (untact) technology. A preprocessing filter and optimal classification model are proposed to improve the accuracy and reliability of heart rate data measured by a non-contact Doppler radar sensor, and the results are compared to those of a contact heart rate sensor (Holter monitor). The MIT-BIH Arrhythmia Database of PhysioNet are used for learning, and the results from the non-contact sensor and Holter monitor are compared for verification. To train the abnormal heartbeat waveform classification model, (1) an optimal heart rate data separation window size is selected through iterative model comparison and used for data separation, and (2) meaningful indicators of heart rate variability are selected; the data are transformed and applied as model characteristics. The non-contact sensor data are then applied to three filter algorithms, and the accuracy is assessed by comparison with the contact sensor data using the trained abnormal heartbeat waveform classification model. Learning is performed using 12 classification models, and the accuracies of the models are compared. This study demonstrates an effective new method of transfer learning for contact data abnormality detection.

Arrhythmia is one of the most persistent chronic heart diseases in the elderly and is associated with high morbidity and mortality such as stroke, cardiac failure, and coronary artery diseases. It is significant for patients with arrhythmias to automatically detect and classify arrhythmia heartbeats using electrocardiogram (ECG) signals. In this paper, we develop three robust deep convolutional neural network (DCNN) models, including a plain-CNN network and two MSF(multi-scale fusion)-CNN architectures (A and B), to aid in better feature extraction for the detection of arrhythmia and thus significantly improve the performance metrics. The proposed models are trained and tested with a public MIT-BIH arrhythmia database on five types of signals. Six groups of ablation experiments are conducted to analyze the performance of the models. The accuracy, sensitivity, and specificity obtained from MSF-CNN architecture A are higher than those from the plain-CNN model, demonstrating that the different parallel group convolution blocks (1 × 3, 1 × 5, and 1 × 7) dramatically improve a model's performance. Additionally, the best model MSF-CNN architecture B achieves an average accuracy, sensitivity, and specificity of 98.00%, 96.17%, and 96.38%, respectively. This illustrates the method with residual learning and concatenation group convolution blocks has a profound effect on the feature learning of the model. The results of ablation experiments show that our proposed biometric recognition and diagnosis network with residual learning (MSF-CNN B) achieves a rapid and reliable diagnosis approach on ECG signal classification, which has the potential for introduction into clinical practice as an excellent tool for aiding cardiologists in reading ECG heartbeat signals.

Arrhythmia is less frequent than a normal heartbeat in an electrocardiogram signal, and the analysis of an electrocardiogram measurement can require more than 24 hours. Therefore, the efficient storage and transmission of electrocardiogram signals have been studied, and their importance has increased recently due to the miniaturization and weight reduction of measurement equipment. The polygonal approximation method based on dynamic programming can effectively achieve signal compression and fiducial point detection by expressing signals with a small number of vertices. However, the execution time and memory area rapidly increase depending on the length of the signal and number of vertices, which are not suitable for lightweight and miniaturized equipment. In this paper, we propose a method that can be applied in embedded environments by optimizing the processing time and memory usage of dynamic programming applied to the polygonal approximation of an ECG signal. The proposed method is divided into three steps to optimize the processing time and memory usage of dynamic programming. The first optimization step is based on the characteristics of electrocardiogram signals in the polygonal approximation. Second, the size of a data bit is used as the threshold for the time difference of each vertex. Finally, a type conversion and memory optimization are applied, which allow real-time processing in embedded environments. After analyzing the performance of the proposed algorithm for a signal length L and number of vertices N, the execution time is reduced from O(L2N) to O(L), and the memory usage is reduced from O(L2N) to O(LN). In addition, the proposed method preserve a performance of fiducial point detection. In a QT-DB experiment provided by Physionet, achieving values of -4.01 ± 7.99 ms and -5.46 ± 8.03 ms.

Vital-sign estimation using ultra-wideband (UWB) radar is preferable because it is contactless and less privacy-invasive. Recently, many approaches have been proposed for estimating heart rate from UWB radar data. However, their performance is still not reliable enough for practical applications. To improve the accuracy, this study employs convolutional neural networks to learn the special patterns of the heartbeats. In the proposed system, skin displacements of the target person are measured using UWB radar, and the radar signal is converted to a two-dimensional matrix, which is used as the input of the designed neural networks. Meanwhile, two triangular waves corresponding to the peaks and valleys in an electrocardiogram are adopted as the output of the networks. The proposed system then identifies each individual and estimates the heart rate automatically based on the already trained neural networks. The estimation error of the interbeat interval computed using our approach was reduced to 4.5 ms in the best case; and 48.5 ms in the worst case. Experiment results show that the proposed approach significantly outperforms a conventional method. The proposed machine learning approach achieves both personal identification and heart rate estimation simultaneously using UWB radar data for the first time. Moreover, this study found that using the respiration and heartbeat components together may enhance the accuracy of heart rate estimation, which is counter-intuitive, because the respiration is usually believed to interfere with the heartbeat.

Physiological and pathological information within electrocardiogram (ECG) is crucial for the diagnosis of heart diseases. Computer-aided diagnosis for the ECG signals has drawn growing research attention up to date. Automatic ECG analysis mainly includes signal denoising, wave detection, and heartbeat classification. These three issues are relevant that the signal denoising can help attenuate the noises and accentuate the typical waves in ECG signals for wave detection, and wave detection can help locate the typical ECG waves and acquire the diagnostically valuable heartbeats based on these waves for the heartbeat classification. The wavelet-based methods play important roles in the three issues, but these methods are scattered and unorganized in the literature. In order to manifest the value of these methods, this paper contributes an overview and taxonomy on them. This paper does the comprehensive summary and systematic categorization on the methods for signal denoising, wave detection, and heartbeat classification according to the deep analysis of their methodological characteristics. By doing so, this paper not only uncovers the inner mechanism that why wavelet-based methods are suitable for ECG analysis but also reveals the designing principles that these methods potentially follow. Finally, this paper has provided an outlook for the developing prospect of “wavelets for ECG” in the future.

Long-term Electrocardiogram (ECG) analysis has become a common means of diagnosing cardiovascular diseases. In order to reduce the workload of cardiologists and accelerate diagnosis, an automated patient-specific heartbeat classification method based on a customized convolutional neural network (CNN) is proposed in this paper. The parallel convolutional layers with kernels of different receptive fields in the network are responsible for extracting multi-spatial deep features of the heartbeats, and the channel-wise attention module is adopted to selectively emphasize the informative features, which are beneficial to distinguish different classes of beats. To facilitate the extraction and emphasis of the important features, each heartbeat is segmented and stacked to form the multi-channel network input according to the basic temporal characteristics of the main components (P wave, QRS complex, and T wave) in the ECG. Besides, to further improve the network generalization and achieve better performance on various ECG of new patients, a method of intra-record sample clustering is proposed to select the representative heartbeats to construct the training set. The proposed method classifies heartbeats into five classes (normal beat (N), supraventricular ectopic beat (SVEB), ventricular ectopic beat (VEB), fusion beat (F), and unclassifiable beat (Q)). Validated on the MIT-BIH arrhythmia database, our approach demonstrates performance superior to several state-of-the-art methods. In addition to an average accuracy and specificity of over 99%, this method achieves a sensitivity of 95.4% and a positive predictivity of 97.1% for VEB class, and a sensitivity of 81.1% and a positive predictivity of 90.0% for SVEB class. With high classification performance and pathological heartbeat detection accuracy, the proposed method is promising for clinical device applications.

Atrial fibrillation (AF), a common abnormal heartbeat rhythm, is a life-threatening recurrent disease that affects older adults. Automatic classification is one of the most valuable topics in medical sciences and bioinformatics, especially the detection of atrial fibrillation. However, it is difficult to accurately explain the local characteristics of electrocardiogram (ECG) signals by manual analysis, due to their small amplitude and short duration, coupled with the complexity and non-linearity. Hence, in this paper, we propose a novel deep arrhythmia-diagnosis method, named deep CNN-BLSTM network model, to automatically detect the AF heartbeats using the ECG signals. The model mainly consists of four convolution layers: two BLSTM layers and two fully connected layers. The datasets of RR intervals (called set A) and heartbeat sequences (P-QRS-T waves, called set B) are fed into the above-mentioned model. Most importantly, our proposed approach achieved favorable performances with an accuracy of 99.94% and 98.63% in the training and validation set of set A, respectively. In the testing set (unseen data sets), we obtained an accuracy of 96.59%, a sensitivity of 99.93%, and a specificity of 97.03%. To the best of our knowledge, the algorithm we proposed has shown excellent results compared to many state-of-art researches, which provides a new solution for the AF automatic detection.

A hatching egg activity classification method aims to accurately and quickly distinguish between dead embryos and live embryos. The existing embryonic classification models collect egg images via a specific imaging system. The image features are then extracted to identify and classify the properties of the hatching eggs. The current state-of-the-art embryonic image classification methods are easily affected by the image quality and are not efficient. To address these issues, we propose a new classification model based on fully convolutional networks (FCNs) and a gated recurrent unit (GRU) that decides whether an embryo is dead or alive by determining embryotic heartbeat signal indicators. Our dataset consists of heartbeat signals from 50k distinct chicken embryos. The experimental results based on our dataset show that our proposed model is the most accurate compared with all baseline models. The reason for this is that our model can capture more useful information from heartbeat signals. In addition, our model can classify 83 hatching eggs per second.

Myocardial infarction (MI) is an acute disease. Early detection and early treatment are of great significance for improving the health of people. In order to reduce the misdiagnosis rate of MI diseases, this paper proposes a multi-lead bidirectional gated recurrent unit neural network (ML-BiGRU) learning algorithm based on current research status in the field of intelligent medical diagnosis, combined with the timing and multi-lead correlation characteristics of the electrocardiogram (ECG) signals. At first, the original ECG signal is denoised and preprocessed and then segmented into heartbeats. After that, the heartbeat sequence is sent to the deep neural network training model to learn the classification. Lastly, the Physikalisch-Technische Bundesanstalt (PTB) ECG database is used to verify the multi-lead BiGRU algorithm. The verification results demonstrate that the accuracy of the algorithm for MI localization is 99.84%, which outperform the other algorithms. The experimental results also show that the algorithm is obviously superior to the traditional localization algorithm in improving the localization accuracy, which is of great significance for improving the correct diagnosis rate of MI.

Electrocardiogram (ECG) signal represents the electrical activity of the heart and playing an increasingly important role for practitioners to diagnose heart diseases. Widely available ECG data and machine learning algorithms present an opportunity to improve the accuracy of automated arrhythmia diagnosis. However, a comprehensive evaluation of morphological arrhythmias for the ECG analysis across a wide variety of diagnostic classes is still a complex task. This paper presents a generic morphological arrhythmias classification method designed for robust and accurate detection of the ECG heartbeat. In this method, the gray-level co-occurrence matrix (GLCM) is employed for features vector description because of its extraordinary statistical feature extraction ability. In addition, the convolutional neural network (CNN) approach is utilized to automatically classification from the generated 3D multi-scale GLCM. Obtained results from the experiments demonstrate that the proposed method in this paper is quite suitable for morphological arrhythmias detection. These findings demonstrate that the GLCM description can efficiently extract the shape features vector for a broad range of distinct arrhythmias from the lead ECGs with high diagnostic performance. The experimental results of the recognition for morphological arrhythmias show the feasibility and effectiveness of the proposed method and could be used to reduce the rate of the misdiagnosed computerized ECG interpretations.

The ultra-wide band (UWB) radar sensor is useful in the field of mission critical sensors and sensor networks due to its short detection time, high penetration, and low energy consumption. To support critical missions such as search and rescue, this paper designs a non-contact detection system based on a UWB radar sensor module to obtain vital signs of human beings. There are noise and clutter due to non-contact detection, and therefore the coherent background noise removal and moving target display filter are applied. Then, the variational mode decomposition (VMD) algorithm is used to extract heartbeat signals and respiratory signals. In addition, the Hilbert transform is applied to heartbeat signals and respiratory signals to obtain the time-frequency information. The electrocardiogram is also employed to compare the results of the UWB radar sensor. We also detect the human target through-the-wall. It turns out that the system can obtain respiratory and heartbeat characteristics simultaneously in one measurement, saving cost with high accuracy.

This paper discusses the use of a nonlinear sensing technology based on radio frequency (RF) intermodulation response to track both the vital signs and location of human subjects. Smart health sensing was realized through the use of a wearable nonlinear tag and an intermodulation-based nonlinear sensor operating in both Doppler and frequency shift keying (FSK) modes. The Doppler mode was used to detect the heartbeat and breathing of the target subject while human subject localization was achieved in the FSK mode. One of the key advantages of this nonlinear smart sensor system was clutter rejection. This system identified the signal reflected from the wearable nonlinear tag and suppressed undesired signals and interferences that were reflected from other objects. The wearable tags used for the experiments were passive, hence they did not require any battery or power supply for their operation. Since the respiration signal is typically stronger than the heartbeat signal, the nonlinear detection setup was designed such that the respiratory signal receives less gain to avoid its sidelobes and harmonics from interfering heartbeat signal detection. This enhanced the heartbeat signal quality so that the cardiac activity could be easily tracked. Four types of experiments were performed on multiple subjects to demonstrate the advantages of this intermodulation-based nonlinear smart health sensing system. Previously, 2nd order harmonics were utilized for target localization and vital sign monitoring. However, these 2nd order harmonics suffer from high path loss and licensing issues. In this paper, target localization and smart health sensing were realized using 3rd order intermodulation with less path loss and no licensing issues compared with its harmonic counterparts. The experiment performed in nonlinear FSK mode was able to detect and locate the source of motion with high accuracy. Similarly, vital signs were recorded in the nonlinear Doppler mode. The design effectively made the amplitude of the heartbeat signal component more prominent, so that the sidelobes and harmonics of respiration do not suppress heartbeat signal.

This paper presents a neural network based processor with improved computation efficiency, which aims at multiclass heartbeat recognition in wearable devices. A lightweight classification algorithm that integrates both bi-directional long short-term memory (BLSTM) and convolutional neural networks (CNN) is proposed to deliver high accuracy with minimal network scale. To reduce energy consumption of the classification algorithm, the similarity between consecutive heartbeats is exploited to achieve a high degree of computation reuse in hardware architecture. In addition, neural network compression techniques are adopted in the procedure of inference to save hardware resources. Synthesized in the SMIC 40LL CMOS process, the prototype design has a total area of 1.40 mm2 with 186.2 kB of static random-access memory (SRAM) capacity. Based on the simulation, this processor achieves an average energy efficiency of 3.52 GOPS/mW under 1.1 V supply at 100 MHz frequency. Compared with the design without computation reuse, the proposed processor provides a speedup by 2.58x and an energy dissipation reduction by 61.27% per classification. This work is a valuable exploration of neural network based design for long-term arrhythmia monitoring in daily life.

Arrhythmia is a disease that threatens human life. Therefore, timely diagnosis of arrhythmia is of great significance in preventing heart disease and sudden cardiac death. The BiLSTM-Attention neural network model with heartbeat activity's global sequence features can effectively improve the accuracy of heartbeat classification. Firstly, the noise is removed by the continuous wavelet transform method. Secondly, the peak of the R wave is detected by the tagged database, and then the P-QRS-T wave morphology and the RR interval are extracted. This feature set is heartbeat activity's global sequence features, which combines single heartbeat morphology and 21 consecutive RR intervals. Finally, the Bi-LSTM algorithm and the BiLSTM-Attention algorithm are used to identify heartbeat category respectively, and the MIT-BIH arrhythmia database is used to verify the algorithm. The results show that the BiLSTM-Attention model combined with heartbeat activity's global sequence features has higher interpretability than other methods discussed in this paper.

Arrhythmias reflect electrical abnormalities of the heart, and they can lead to severe harm to the heart. An electrocardiogram (ECG) is a useful tool to manifest arrhythmias. In this paper, we present an automatic system using a convolutional neural network and active learning to classify ECG signals. To improve the model performance, breaking-ties (BT) and modified BT algorithms are utilized in the active learning. We classify ECG signals in five heartbeat types, i.e., normal (N), ventricular (V), supraventricular (S), fusion of normal and ventricular (F), and unknown heartbeats (Q), using the Association for the Advancement of Medical Instrumentation standard. Our experiments are performed on the MIT-BIH arrhythmia database. To further verify the generalization capability of the system, the ECG data that acquired from our wearable device are also used to conduct in the experiments. Compared with most of the state-of-the-art methods, the obtained results demonstrate that the presented method promotes the classification performance remarkably.

A high-performance method of automated heartbeat classification based on Convolutional Neural Network (CNN) is proposed in this paper. To make full use of the electrocardiogram information acquired from different parts of the human body, we present a novel 3-D data structure as the input of the CNN. The 3-D structure consists of multiple feature maps, each of which indicates the information collected from one lead and contains morphological characteristic, RR-interval, and beat-to-beat correlation feature. Besides, atrous spatial pyramid pooling (ASPP) module which uses filters with different resolutions is adopted to extract deep features in multi-fields of view. Validated on the MIT-BIH arrhythmia database, the proposed method yields an overall accuracy of 91.44% in the inter-patient practice. In particular, this method achieves 89.05% and 95.15% in the sensitivities of the supraventricular ectopic beat (SVEB) and ventricular ectopic beat (VEB) classes, respectively. With the high performance in detecting these two pathological classes, this method has potential clinical application.

Correct heartbeat classification from electrocardiogram (ECG) signals is fundamental to the diagnosis of arrhythmia. The recent advancement in deep convolutional neural network (CNN) has renewed the interest in applying deep learning techniques to improve the accuracy of heartbeat classification. So far, the results are not very exciting. Most of the existing methods are based on ECG morphological information, which makes deep learning difficult to extract discriminative features for classification. Towards an opposite direction of feature extraction or selection, this paper proceeds along a recent proposed direction named feature enrichment (FE). To exploit the advantage of deep learning, we develop a FE-CNN classifier by enriching the ECG signals into time-frequency images by discrete short-time Fourier transform and then using the images as the input to CNN. Experiments on MIT-BIH arrhythmia database show FE-CNN obtains sensitivity (Sen) of 75.6%, positive predictive rate (Ppr) of 90.1%, and F1 score of 0.82 for the detection of supraventricular ectopic (S) beats. Sen, Ppr, and F1 score are 92.8%, 94.5%, and 0.94, respectively, for ventricular ectopic (V) beat detection. The result demonstrates our method outperforms state-of-the-art algorithms including other CNN based methods, without any hand-crafted features, especially F1 score for S beat detection from 0.75 to 0.82. This FE-CNN classifier is simple, effective, and easy to be applied to other types of vital signs.

Impulse radio ultrawideband radar is a critical remote sensing tool for life detection and noncontact monitoring of vital signals. In noncontact monitoring via radar, the disturbance from respiration and environmental noise is considered critical for the estimation of heart rates. However, the heartbeat signal is generally distorted by breath harmonics and fluctuations in the time domain, and the frequencies of the vital signals are closely situated; thus, it is difficult to employ an ordinary frequency filter for separation. To solve this problem, a novel method was developed to extract heartbeat information. In this study, convolutional sparse coding, which is an unsupervised machine learning algorithm, was first used to model the heartbeat signal in the time domain, given the respiration and relative artifacts. The proposed scheme was then used to decompose the time-domain signals and directly obtain the heartbeat component by exploiting the sparsity of the heartbeat signal in the time domain. Furthermore, the performance of the proposed scheme was improved using a noise-assisted method, and the process was accelerated by learning from the K-singular value decomposition strategy. Finally, by testing the vital sign signals collected from the finite differences time-domain simulation and experiments, the results obtained indicate that the proposed approach is effective for the extraction of low-amplitude heartbeat signals from the respiration signal, and that it significantly improves the accuracy of heart rate evaluation.

A high performance electrocardiogram (ECG)-based arrhythmic beats classification system is presented in this paper. The classifier was designed based on convolutional neural network (CNN). Single channel ECG signal was segmented into heartbeats in accordance with the changing heartbeat rate. The beats were transformed into dual beat coupling matrix as 2-D inputs to the CNN classifier, which captured both beat morphology and beat-to-beat correlation in ECG. A systematic training beat selection procedure was also proposed which automatically include the most representative beats into the training set to improve classification performance. The classification system was evaluated for the detection of supraventricular ectopic beats (SVEB or S beats) and VEB using the MIT-BIH arrhythmia database. Our proposed method has demonstrated superior performance than several state-of-the-art detectors. In particular, our proposed CNN system has improved sensitivity and positive predictive rate for S beats by more than 12.2% and 11.9%, respectively, over these top performing algorithms. Our proposed CNN classifier with an automatic training beats selection process has shown to outperform the previous methods. The classifier is also a personalized one by combining training set from a common pool and a subject-specific set of ECG data. Our proposed system provides a reliable and fully automatic tool for detection of arrhythmia heartbeat without the need for manual feature extraction or expert assistant. It can potentially be implemented on portable device for the long-term monitoring of cardiac arrhythmia.

In this paper, we propose a fully ear-worn long-term blood pressure (BP) and heart rate (HR) monitor to achieve a higher wearability. Moreover, to enable practical application scenarios, we present a machine learning framework to deal with severe motion artifacts induced by head movements. We suggest situating all electrocardiogram (ECG) and photoplethysmography (PPG) sensors behind two ears to achieve a super wearability, and successfully acquire weak ear-ECG/PPG signals using a semi-customized platform. After introducing head motions toward real-world application scenarios, we apply a support vector machine classifier to learn and identify raw heartbeats from motion artifacts-impacted signals. Furthermore, we propose an unsupervised learning algorithm to automatically filter out residual distorted/faking heartbeats, for ECG-to-PPG pulse transit time (PTT) and HR estimation. Specifically, we introduce a dynamic time warping-based learning approach to quantify distortion conditions of raw heartbeats referring to a high-quality heartbeat pattern, which are then compared with a threshold to perform purification. The heartbeat pattern and the distortion threshold are learned by a K-medoids clustering approach and a histogram triangle method, respectively. Afterward, we perform a comparative analysis on ten PTT or PTT&HR-based BP learning models. Based on an acquired data set, the BP and HR estimation using the proposed algorithm has an error of -1.4±5.2 mmHg and 0.8±2.7 beats/min, respectively, both much lower than the state-of-the-art approaches. These results demonstrate the capability of the proposed machine learning-empowered system in ear-ECG/PPG acquisition and motion-tolerant BP/HR estimation. This proof-of-concept system is expected to illustrate the feasibility of ear-ECG/PPG-based motion-tolerant BP/HR monitoring.

Electrocardiogram (ECG) signals from mobile sensors are expected to increase the availability of authentication in the emerging wearable device industry. However, mobile sensors provide a relatively lower quality signal than the conventional medical devices. This paper proposes a practical authentication procedure for ECG signals that collected via one-chip-solution mobile sensors. We designed a cascading bandpass filter for noise cancellation and suggest eight fiducial features. For classification-based authentication, we use the radial basis function kernel-based support vector machine showing the best performance among nine classifiers through experimental comparisons. In spite of noisy ECG signals in mobile sensors, we achieved 4.61% of the equal error rate (EER) on a single heartbeat, and 1.87% of EER on 15 s testing time on 175 subjects, which is a reasonable result and supports the usability of low-cost ECGs for biometric authentication.