Electrocardiogram Time Series Research Articles

A novel hybrid GRU-CNN and residual bias (RB) based RB-GRU-CNN models for prediction of PTB Diagnostic ECG time series data

Time series analysis is crucial in the field of medical sciences. Recently, it has become crucial to comprehend the intricacies of data to forecast and assess the likelihood of survival in patients with different ailments. The study employs diverse datasets from multiple PTB electrocardiogram (ECG) datasets, which exhibit a high degree of uncorrelation. These datasets include patients with various heart diseases and age groups, ensuring dataset diversity. The paper presents two models that utilize deep learning and residual bias. The primary approach proposed is a hybrid GRU-CNN model. In contrast, the second model incorporates residual bias (RB) to improve upon the first model, referred to as the RB-GRU-CNN model. The hybrid GRU-CNN model has been compared to conventional deep learning models across multiple metrics, achieving the highest average percentage of improvement in RMSE (46.73%), MSE (67.46%), MAPE (43.92%), and MAE (48.52%) compared to LSTM. The RB-GRU-CNN proposed model demonstrated greater efficacy than the GRU-CNN proposed model, as confirmed by non-parametric statistical Friedman ranking and their corresponding Holm’s p-values. The comprehensive study of the results proves that the proposed models outperformed the performance of the traditional models.

Intelligent fatigue detection based on hierarchical multi-scale ECG representations and HRV measures

Fatigue detection based on electrocardiogram (ECG) from wearable devices is a convenient approach during exercise. However, how to model the data variations of ECG time series and capture effective information for accurate fatigue detection is still challenging. In this study, we propose a novel fatigue assessment architecture designed to adaptively extract and integrate both the ECG and heart rate variability (HRV) features, so as to obtain rich contextual information from ECG sequences and improve exercise-induced fatigue prediction performance. Inspired by the ECG observation process of human experts, the hierarchical transformer framework is designed to model the long-range dependencies of ECG time series and obtain multi-scale ECG representations. To reduce the computation cost, we also devise a sampling-efficient multi-head attention (SMA) mechanism. Finally, a novel gated feature fusion layer is developed to adaptively learn the weights of both the ECG and HRV features for fatigue prediction. The proposed fatigue prediction system is validated with real fatigue datasets which were collected from healthy participants. The experimental results show that the proposed approach outperforms other recent state-of-the-art systems in terms of accuracy 94.0%, specificity 99.0%, recall 89.3%, F1-score 93.7%, G-mean score 91.7%, MCC 0.837, and Cohen’s Kappa 0.847, based on the ECG time series with duration 20s. This demonstrates that the proposed system can be used to accurately detect fatigue in real time during exercise, such that assisting coaches or athletes to monitor and adjust their exercise session for preventing overexertion and fatigue-induced injury.

Enhancing ECG classification with continuous wavelet transform and multi-branch transformer

BackgroundAccurate classification of electrocardiogram (ECG) signals is crucial for automatic diagnosis of heart diseases. However, existing ECG classification methods often require complex preprocessing and denoising operations, and traditional convolutional neural network (CNN)-based methods struggle to capture complex relationships and high-level time-series features. MethodIn this study, we propose an ECG classification method based on continuous wavelet transform and multi-branch transformer. The method utilizes continuous wavelet transform (CWT) to convert the ECG signal into time-series feature map, eliminating the need for complicated preprocessing. Additionally, the multi-branch transformer is introduced to enhance feature extraction during model training and improve classification performance by removing redundant information while preserving important features. ResultsThe proposed method was evaluated on the CPSC 2018 (6877 cases) and MIT-BIH (47 cases) ECG public datasets, achieving an accuracy of 98.53% and 99.38%, respectively, with F1 scores of 97.57% and 98.65%. These results outperformed most existing methods, demonstrating the excellent performance of the proposed method. ConclusionThe proposed method accurately classifies the ECG time-series feature map, which holds promise for the diagnosis of cardiac arrhythmias. The findings of this study are valuable for advancing the field of automatic ECG diagnosis.

Tetanus Severity Classification in Low-Middle Income Countries through ECG Wearable Sensors and a 1D-Vision Transformer

Tetanus, a life-threatening bacterial infection prevalent in low- and middle-income countries like Vietnam, impacts the nervous system, causing muscle stiffness and spasms. Severe tetanus often involves dysfunction of the autonomic nervous system (ANS). Timely detection and effective ANS dysfunction management require continuous vital sign monitoring, traditionally performed using bedside monitors. However, wearable electrocardiogram (ECG) sensors offer a more cost-effective and user-friendly alternative. While machine learning-based ECG analysis can aid in tetanus severity classification, existing methods are excessively time-consuming. Our previous studies have investigated the improvement of tetanus severity classification using ECG time series imaging. In this study, our aim is to explore an alternative method using ECG data without relying on time series imaging as an input, with the aim of achieving comparable or improved performance. To address this, we propose a novel approach using a 1D-Vision Transformer, a pioneering method for classifying tetanus severity by extracting crucial global information from 1D ECG signals. Compared to 1D-CNN, 2D-CNN, and 2D-CNN + Dual Attention, our model achieves better results, boasting an F1 score of 0.77 ± 0.06, precision of 0.70 ± 0. 09, recall of 0.89 ± 0.13, specificity of 0.78 ± 0.12, accuracy of 0.82 ± 0.06 and AUC of 0.84 ± 0.05.

Machine learning-based cardiac activity non-linear analysis for discriminating COVID-19 patients with different degrees of severity

Objective:This study highlights the potential of an Electrocardiogram (ECG) as a powerful tool for early diagnosis of COVID-19 in critically ill patients with limited access to CT–Scan rooms. Methods:In this investigation, 3 categories of patient status were considered: Low, Moderate, and Severe. For each patient, 2 different body positions have been used to collect 2 ECG signals. Then, from each collected signal, 10 non-linear features (Energy, Approximate Entropy, Logarithmic Entropy, Shannon Entropy, Hurst Exponent, Lyapunov Exponent, Higuchi Fractal Dimension, Katz Fractal Dimension, Correlation Dimension and Detrended Fluctuation Analysis) were extracted every 1s ECG time-series length to serve as entries for 19 Machine learning classifiers within a leave-one-out cross-validation procedure. Four different classification scenarios were tested: Low vs. Moderate, Low vs. Severe, Moderate vs. Severe and one Multi-class comparison (All vs. All). Results:The classification report results were: (1) Low vs. Moderate - 100% of Accuracy and 100% of F1–Score; (2) Low vs. Severe - Accuracy of 91.67% and an F1–Score of 94.92%; (3) Moderate vs. Severe - Accuracy of 94.12% and an F1–Score of 96.43%; and (4) All vs All - 78.57% of Accuracy and 84.75% of F1–Score. Conclusion:The results indicate that the applied methodology could be considered a good tool for distinguishing COVID-19’s different severity stages using ECG signals. Significance:The findings highlight the potential of ECG as a fast and effective tool for COVID-19 examination. In comparison to previous studies using the same database, this study shows a 7.57% improvement in diagnostic accuracy for the All vs All comparison.

Data programming enabled weak supervised labeling for ECG time series

Electrocardiogram (ECG) beat labeling performed using conventional methods is unsuitable for ECG signals obtained from Internet of Things (IoT) wearable devices. The conventional methods employ manually labeled data captured using multiple leads, while most IoT devices produce unlabeled single lead data. Getting ECG data labeled by a subject matter expert (SME) is a resource/time/cost-intensive task. Our research addresses this challenge by proposing an automatic labeling technique for ECG time-series data obtained from a single lead. The technique employs a data programming (DP) enabled weak supervised learning (WSL) technique for automatic labeling of ECG beats. We have proposed nine novel heuristics-based labeling functions (LFs), applied them to each ECG beat and subsequently used a generative model (GM) to assign a probabilistic label to each ECG beat employing both intra and inter-patient paradigm on MIT-BIH and INCART datasets. Further, a discriminative model (DM) is trained on top of the GM for maximizing data coverage and ensure robustness. Finally, data augmentation (DA) is used to solve the class imbalance problem inherent in ECG data. Our experimental results demonstrate a simpler, faster, and more accurate labeling method i.e., approximately 105 ECG beats are labeled in an hour with an accuracy of 92.2% from single lead data. In contrast with human annotators, the time and cost requirements of our proposed labeling method are significantly less.

Utilizing variable auto encoder-based TDO optimization algorithm for predicting loneliness from electrocardiogram signals.

Several seniors and a substantial part of the general population are living in social isolation. This frequently occurs in vulnerability, isolation, and depression, which then have a poor impact on other health-related factors. A number of health problems, including a higher risk of cardio problems, are brought on by social isolation and loneliness. Electrocardiogram (ECG) usage for mental condition recognition enables accurate determination of a person's internal representation. The electrocardiogram (ECG) signals can be thoroughly analyzed to uncover hidden data that may be helpful for the precise identification of cardiac problems. ECG time-series information typically have great dimensions and complicated componentry. Using relevant information to guide training is among the main achievements of this type of learning. An ECG signal plays a significant part in the individual body's ability to manage behavior. Furthermore, loneliness identification is crucial since it has the worse effect on the circumstances that afflict persons. This study suggested an approach for detecting loneliness from an ECG signal to use a variable auto encoder-based optimization algorithm for ESN technique. The suggested approach consists of three phases for identifying a person's loneliness. Firstly, undecimated discrete wavelet transform is used to preprocess the acquired ECG data. Next, further characteristics are extracted from the precompiled signals using a variable auto encoder. For the precise categorization of loneliness in the ECG signal, a metaheuristic optimized ESN is, therefore, presented. The outcomes of the tests demonstrate that the suggested system with suitable ECG representations produces improved accuracy as well as performance.

Unsupervised Transformer-Based Anomaly Detection in ECG Signals

Anomaly detection is one of the basic issues in data processing that addresses different problems in healthcare sensory data. Technology has made it easier to collect large and highly variant time series data; however, complex predictive analysis models are required to ensure consistency and reliability. With the rise in the size and dimensionality of collected data, deep learning techniques, such as autoencoder (AE), recurrent neural networks (RNN), and long short-term memory (LSTM), have gained more attention and are recognized as state-of-the-art anomaly detection techniques. Recently, developments in transformer-based architecture have been proposed as an improved attention-based knowledge representation scheme. We present an unsupervised transformer-based method to evaluate and detect anomalies in electrocardiogram (ECG) signals. The model architecture comprises two parts: an embedding layer and a standard transformer encoder. We introduce, implement, test, and validate our model in two well-known datasets: ECG5000 and MIT-BIH Arrhythmia. Anomalies are detected based on loss function results between real and predicted ECG time series sequences. We found that the use of a transformer encoder as an alternative model for anomaly detection enables better performance in ECG time series data. The suggested model has a remarkable ability to detect anomalies in ECG signal and outperforms deep learning approaches found in the literature on both datasets. In the ECG5000 dataset, the model can detect anomalies with 99% accuracy, 99% F1-score, 99% AUC score, 98.1% recall, and 100% precision. In the MIT-BIH Arrhythmia dataset, the model achieved an accuracy of 89.5%, F1 score of 92.3%, AUC score of 93%, recall of 98.2%, and precision of 87.1%.

Cloud-based healthcare framework for real-time anomaly detection and classification of 1-D ECG signals.

Real-time data collection and pre-processing have enabled the recognition, realization, and prediction of diseases by extracting and analysing the important features of physiological data. In this research, an intelligent end-to-end system for anomaly detection and classification of raw, one-dimensional (1D) electrocardiogram (ECG) signals is given to assess cardiovascular activity automatically. The acquired raw ECG data is pre-processed carefully before storing it in the cloud, and then deeply analyzed for anomaly detection. A deep learning-based auto-encoder(AE) algorithm is applied for the anomaly detection of 1D ECG time-series signals. As a next step, the implemented system identifies it by a multi-label classification algorithm. To improve the classification accuracy and model robustness the improved feature-engineered parameters of the large and diverse datasets have been incorporated. The training has been done using the amazon web service (AWS) machine learning services and cloud-based storage for a unified solution. Multi-class classification of raw ECG signals is challenging due to a large number of possible label combinations and noise susceptibility. To overcome this problem, a performance comparison of a large set of machine algorithms in terms of classification accuracy is presented on an improved feature-engineered dataset. The proposed system reduces the raw signal size up to 95% using wavelet time scattering features to make it less compute-intensive. The results show that among several state-of-the-art techniques, the long short-term memory (LSTM) method has shown 100% classification accuracy, and an F1 score on the three-class test dataset. The ECG signal anomaly detection algorithm shows 98% accuracy using deep LSTM auto-encoders with a reconstructed error threshold of 0.02 in terms of absolute error loss. Our approach provides performance and predictive improvement with an average mean absolute error loss of 0.0072 for normal signals and 0.078 for anomalous signals.

Abstract 14935: Deep Learning Pipeline for Frailty Screening Using the 12-Lead Electrocardiogram

Introduction: The electrocardiogram (ECG) contains information about age-related changes in cardiovascular physiology, which have been linked with the frailty syndrome. Hypothesis: We sought to develop and validate a predictive model leveraging the 12-lead ECG to screen for frailty as defined by a prospective reference standard. Methods: We conducted a population-based cohort study using data from the Canadian Longitudinal Study of Aging (CLSA). From 2010-2015, the CLSA enlisted a diverse and multi-ethnic sample of community-dwelling adults 45-85 years of age. Comprehensive phenotyping was performed through interviews at participants’ homes and assessments at data collection sites. Frailty was quantified by the 110-item Frailty Index (FI) Composite, consisting of self-reported comorbidities, blood tests, physical performance tests, body composition tests, cardiovascular and pulmonary tests, cognitive and sensory tests. After dividing our sample into training (80%) and test (20%) sets, we developed an end-to-end deep neural network to predict the FI score based on the 12-lead ECG time series. Results: A total of 26,700 ECGs with paired FI scores were evaluated. For classification of FI quintiles, a bidirectional long short-term memory (BiLSTM) neural network with a cross-entropy loss function achieved a 5-fold mean area under the receiver operating characteristics curve (AUROC) of 0.70 and area under the precision-recall curve (AUPRC) of 0.36. Predictive performance was superior for classification of the first (most robust) quintile that had AUROC 0.79 and AUPRC 0.46, and the fifth (most frail) quintile that had AUROC 0.79 and AUPRC 0.56, as compared to the middle quintiles that had AUROC 0.60-0.69 and AUPRC 0.27-0.29. Conclusions: Our deep learning model can be used to screen for high or low levels of frailty based on the readily available 12-lead ECG. Additional research is underway to gain insights into other representations of the ECG signal and relative importance of the ECG features.

Electrocardiogram Signal Classification Based on Mix Time-Series Imaging

Arrhythmia is a significant cause of death, and it is essential to analyze the electrocardiogram (ECG) signals as this is usually used to diagnose arrhythmia. However, the traditional time series classification methods based on ECG ignore the nonlinearity, temporality, or other characteristics inside these signals. This paper proposes an electrocardiogram classification method that encodes one-dimensional ECG signals into the three-channel images, named ECG classification based on Mix Time-series Imaging (EC-MTSI). Specifically, this hybrid transformation method combines Gramian angular field (GAF), recurrent plot (RP), and tiling, preserving the original ECG time series’ time dependence and correlation. We use a variety of neural networks to extract features and perform feature fusion and classification. This retains sufficient details while emphasizing local information. To demonstrate the effectiveness of the EC-MTSI, we conduct abundant experiments in a commonly-used dataset. In our experiments, the general accuracy reached 93.23%, and the accuracy of identifying high-risk arrhythmias of ventricular beats and supraventricular beats alone are as high as 97.4% and 96.3%, respectively. The results reveal that the proposed method significantly outperforms the existing approaches.

Robust Reconstruction of Electrocardiogram Using Photoplethysmography: A Subject-Based Model.

Electrocardiography and photoplethysmography are non-invasive techniques that measure signals from the cardiovascular system. While the cycles of the two measurements are highly correlated, the correlation between the waveforms has rarely been studied. Measuring the photoplethysmogram (PPG) is much easier and more convenient than the electrocardiogram (ECG). Recent research has shown that PPG can be used to reconstruct the ECG, indicating that practitioners can gain a deep understanding of the patients’ cardiovascular health using two physiological signals (PPG and ECG) while measuring only PPG. This study proposes a subject-based deep learning model that reconstructs an ECG using a PPG and is based on the bidirectional long short-term memory model. Because the ECG waveform may vary from subject to subject, this model is subject-specific. The model was tested using 100 records from the MIMIC III database. Of these records, 50 had a circulatory disease. The results show that a long ECG signal could be effectively reconstructed from PPG, which is, to our knowledge, the first attempt in this field. A length of 228 s of ECG was constructed by the model, which was trained and validated using 60 s of PPG and ECG signals. To segment the data, a different approach that segments the data into short time segments of equal length (and that do not rely on beats and beat detection) was investigated. Segmenting the PPG and ECG time series data into equal segments of 1-min width gave the optimal results. This resulted in a high Pearson’s correlation coefficient between the reconstructed 228 s of ECG and referenced ECG of 0.818, while the root mean square error was only 0.083 mV, and the dynamic time warping distance was 2.12 mV per second on average.

Validation of Psychophysiological Measures for Caffeine Oral Films Characterization by Machine Learning Approaches.

(1) Background: The oral films are a new delivery system that can carry several molecules, such as neuromodulator molecules, including caffeine. These delivery systems have been developed and characterized by pharmacokinetics assays. However, new methodologies, such as psychophysiological measures, can complement their characterization. This study presents a new protocol with psychophysiological parameters to characterize the oral film delivery systems based on a caffeine model. (2) Methods: Thirteen volunteers (61.5% females and 38.5% males) consumed caffeine oral films and placebo oral films (in different moments and without knowing the product). Electrocardiogram (ECG), electrodermal (EDA), and respiratory frequency (RF) data were monitored for 45 min. For the data analysis, the MATLAB environment was used to develop the analysis program. The ECG, EDA, and RF signals were digitally filtered and processed, using a windowing process, for feature extraction and an energy mean value for 5 min segments. Then, the data were computed and presented to the entries of a set of Machine Learning algorithms. Finally, a data statistical analysis was carried out using SPSS. (3) Results: Compared with placebo, caffeine oral films led to a significant increase in power energy in the signal spectrum of heart rate, skin conductance, and respiratory activity. In addition, the ECG time-series power energy activity revealed a better capacity to detect caffeine activity over time than the other physiological modalities. There was no significant change for the female or male gender. (4) Conclusions: The protocol developed, and the psychophysiological methodology used to characterize the delivery system profile were efficient to characterize the drug delivery profile of the caffeine. This is a non-invasive, cheap, and easy method to apply, can be used to determine the neuromodulator drugs delivery profile, and can be implemented in the future.

Detecting a broader spectrum of cardiac pathologies in electrocardiogram data by applying a deep neural network designed to detect a specific cardiac disease

Abstract Background/Introduction The electrocardiogram (ECG) is a widely used and inexpensive tool that provides extensive insights into the cardiac structure and function. Artificial intelligence (AI) algorithms, especially deep learning (DL) models, are efficient computer based instruments with which large and complex datasets can be processed for identification of e.g. specific diseases. PhysioNet is a NIH research resource for complex signals including a large amount of labelled ECG time-series data. Our aim was to evaluate the diagnostic performance of an AI architecture developed to detect a specific cardiac pathology in a large ECG data set including a broad range of cardiac abnormalities. Methods The PhysioNet ECG dataset provided as part of the PhysioNet Challenge 2020 consists of five distinct databases with a total of 43100 12-Lead ECG recordings of varying length stemming from patients from China, Russia, Europe and the United States. Each ECG recording is annotated with diagnoses based on a set of 111 possible labels, which express either a cardiac pathology, e.g. atrial flutter or anterior wall ischemia, or unspecific changes in the ECG, e.g. a prolonged qt interval or low qrs voltages. Based on these labels we defined 10 groups merging PhysioNet labels describing related cardiac abnormalities (see Table 1). We adapted a recently published DL model which used raw ECG time-series data of all 12-leads rather than extracted features as model input. This DL model was adapted to the larger number of output variables and then trained on 80% (n=34480 ECGs) of the PhysioNet dataset. The remaining 20% (n=8620 ECGs) of the PhysioNet dataset were used to evaluate the diagnostic performance of the AI model. Sensitivities, specificities and the areas under the receiver operator characteristic curves (AUROC) were used as performance metrices. Results The AI model, that was initially designed to detect a specific cardiac pathology, performed well in the large PhysioNet dataset providing AUROCs ranging from 0.78 to 0.95 to detect the defined 10 cardiac abnormality groups. Interestingly, the AI model was able to detect disease groups with changes in the chronological sequence of the ECG, e.g. arrhythmia, with comparable precision as disease groups associated primarily with changes in the ECG amplitude like e.g. ischemia. Detailed results are presented in Table 2. Conclusion(s) Our evaluation shows that an AI model that uses raw ECG time-series data rather than extracted features as model input can be easily transferred to other large datasets with different prediction variables. This might also serve as a proof of concept that raw data instead of pre-selected features should be used as model input if developing AI applications for medical use cases. Funding Acknowledgement Type of funding sources: Public Institution(s). Main funding source(s): FlexiFunds by Forschungscampus Mittelhessen

Predicting mortality in cardiovascular patients using electrocardiogram data and artificial intelligence

Abstract Background/Introduction The electrocardiogram (ECG) is an ubiquitously used non-invasive tool for diagnosis and risk prediction in cardiology, granting deep extensive insights into the heart. Artificial intelligence (AI) is a modern resource allowing the processing of vast complex datasets in a way that is comparable to humans. Risk stratification in cardiovascular patients is mainly based on scoring systems, such as the ESC-SCORE, relying on traditional risk variables like cholesterol levels or arterial hypertension, rather than actual cardiac structure and function. Goal of this project was to predict mortality using AI in patients with cardiovascular risk based on the current cardiac situation represented by a standard 12-lead ECG recording. Methods The study population is based on an ongoing registry that started in 2010 and enrolled patients scheduled for an invasive coronary angiography due to suspected chronic coronary syndrome. Data of the following study patients were analysed: enrolment within the first two study years with available long-term follow-up data on the outcome measure overall mortality, availability of an ECG at admission without pacemaker stimulation and availability of all variables needed to calculate the ESC-SCORE (in the version weighed for a German population) as comparison. This led to a cohort of 720 patients, of whom 70 died within the follow-up period. Information on presence of a relevant coronary artery disease (CAD) was available for all patients, to differentiate between primary and secondary prevention. A deep learning architecture that was previously developed to detect myocardial scar in raw ECG time-series data was used. This model was trained with 1400 ECG recordings, from the publicly available PTB-XL dataset with 700 of those ECGs labelled for acute, recent or old myocardial infarction while 700 were labelled as healthy. This pre-trained model was then applied to our study cohort to predict long-term mortality based on a single 12-lead ECG obtained at admission. Results For mortality prediction in patients without CAD (primary prevention) the AI model compares to the ESC-SCORE with an AUROC of 0.606 vs 0.584. For CAD patients (secondary prevention) the AI model compares with an AUROC of 0.612 vs 0.658. Detailed results are presented in Table 1. Conclusion(s) Our data underlines the potential of an AI based approach, predicting mortality in cardiovascular patients using only single 12-lead ECG recordings. Additionally, our model achieved similar predictive information to established risk classification systems, such as the ESC-SCORE. Since data acquisition is still ongoing, we will continue to improve our model. In future work training AI to specifically predict mortality while also exploring explainable AI could lead to breakthrough findings in ECG interpretation. Funding Acknowledgement Type of funding sources: Public Institution(s). Main funding source(s): FlexiFunds by Forschungscampus Mittelhessen

MED-NET: a novel approach to ECG anomaly detection using LSTM auto-encoders

Time series data is generated in various sectors of day to day life. Among all, one of the most important areas of generation and processing of time series data plays vital role in medical analysis. In this context, various continuous time series dependent EEG and ECG (electrocardiogram) signals are the most important types of medical signals produced and monitored by doctors. This paper proposes a highly novel and robust approach to analyse and detect ECG signals for tracking of anomalies in the signals using Hybrid Deep Learning Architectures (HDLA). The proposed scheme implements self-supervised pattern recognition using Long Short-Term Memory (LSTM) networks in terms of autoencoder and decoder. Finally, the proposed scheme is tested on Physio-net data set. The outcome of the model can also handle noise associated with ECG time series signal, it achieves high accuracy and also solves overfitting problems in a robust and efficient manner.

The Use of Time-Frequency Moments as Inputs of LSTM Network for ECG Signal Classification

This paper refers to the method of using the deep neural long-short-term memory (LSTM) network for the problem of electrocardiogram (ECG) signal classification. ECG signals contain a lot of subtle information analyzed by doctors to determine the type of heart dysfunction. Due to the large number of signal features that are difficult to identify, raw ECG data is usually not suitable for use in machine learning. The article presents how to transform individual ECG time series into spectral images for which two characteristics are determined, which are instantaneous frequency and spectral entropy. Feature extraction consists of converting the ECG signal into a series of spectral images using short-term Fourier transformation. Then the images were converted using Fourier transform again to two signals, which includes instantaneous frequency and spectral entropy. The data set transformed in this way was used to train the LSTM network. During the experiments, the LSTM networks were trained for both raw and spectrally transformed data. Then, the LSTM networks trained in this way were compared with each other. The obtained results prove that the transformation of input signals into images can be an effective method of improving the quality of classifiers based on deep learning.

Classification of Arrhythmia by Using Deep Learning with 2-D ECG Spectral Image Representation

The electrocardiogram (ECG) is one of the most extensively employed signals used in the diagnosis and prediction of cardiovascular diseases (CVDs). The ECG signals can capture the heart’s rhythmic irregularities, commonly known as arrhythmias. A careful study of ECG signals is crucial for precise diagnoses of patients’ acute and chronic heart conditions. In this study, we propose a two-dimensional (2-D) convolutional neural network (CNN) model for the classification of ECG signals into eight classes; namely, normal beat, premature ventricular contraction beat, paced beat, right bundle branch block beat, left bundle branch block beat, atrial premature contraction beat, ventricular flutter wave beat, and ventricular escape beat. The one-dimensional ECG time series signals are transformed into 2-D spectrograms through short-time Fourier transform. The 2-D CNN model consisting of four convolutional layers and four pooling layers is designed for extracting robust features from the input spectrograms. Our proposed methodology is evaluated on a publicly available MIT-BIH arrhythmia dataset. We achieved a state-of-the-art average classification accuracy of 99.11%, which is better than those of recently reported results in classifying similar types of arrhythmias. The performance is significant in other indices as well, including sensitivity and specificity, which indicates the success of the proposed method.

A Multi-Task Group Bi-LSTM Networks Application on Electrocardiogram Classification.

Background: Cardiovascular diseases (CVD) are the leading cause of death globally. Electrocardiogram (ECG) analysis can provide thoroughly assessment for different CVDs efficiently. We propose a multi-task group bidirectional long short-term memory (MTGBi-LSTM) framework to intelligent recognize multiple CVDs based on multi-lead ECG signals. Methods: This model employs a Group Bi-LSTM (GBi-LSTM) and Residual Group Convolutional Neural Network (Res-GCNN) to learn the dual feature representation of ECG space and time series. GBi-LSTM is divided into Global Bi-LSTM and Intra-Group Bi-LSTM, which can learn the features of each ECG lead and the relationship between leads. Then, through attention mechanism, the different lead information of ECG is integrated to make the model to possess the powerful feature discriminability. Through multi-task learning, the model can fully mine the association information between diseases and obtain more accurate diagnostic results. In addition, we propose a dynamic weighted loss function to better quantify the loss to overcome the imbalance between classes. Results: Based on more than 170,000 clinical 12-lead ECG analysis, the MTGBi-LSTM method achieved accuracy, precision, recall and F1 of 88.86%, 90.67%, 94.19% and 92.39%, respectively. The experimental results show that the proposed MTGBi-LSTM method can reliably realize ECG analysis and provide an effective tool for computer-aided diagnosis of CVD.

The complexity of ECG signal based on multifractal theories and its nonlinear dynamical mechanism

The nonlinear dynamical analysis of biomedical signals has always been the research focus in academic fields. It better uncovers the essential rules in life activities than conventional linear and time-frequency analysis methods. Human body surface electrocardiogram (ECG) signal is non-stationary and frequency-varying by nature, which belongs to a typical nonlinear signal. Therefore, traditional linear analysis methods cannot completely disclose its nonlinear nature. The existing nonlinear dynamic tools, such as correlation dimension (D2), Lyapunov exponent, entropy, detrended fluctuation analysis, multifractal singularity spectrum etc., have already been applied to the studies on heart electrical signals, which proved that this signal has definite chaotic characteristic. The aforementioned nonlinear methods perform better in the analysis of deterministic stochastic signals. In our viewpoints, because all the probabilities in the original data set can be exactly expressed by the corresponding points in the multifractal singularity spectrum f ( α ), the area of the spectrum must comprise the total information of the data. This two-dimensional expression is more accurate than the computation of spectral width Δ α in analyzing the heartbeat signal. With regard to ECG time series, the areas of singularity spectrum should include all the information about the heartbeat dynamics. Considering the irregularity and anisotropy of the heartbeat electrical signal propagation, we investigated the arithmetic mean and dispersing degree of the area calculated from synchronous 12-lead ECG signals obtained from a large number of subjects, who were under different physiological and pathological conditions. The 12-lead ECG records were obtained from 12 sensors positioned on body surfaces, including the limb and chest leads. In addition, the ECG signals were in high frequency section (HFECG), whose frequency components were above 100 Hz (approximately 100 Hz–2 kHz). Some early heart diseases can first be reflected by high-frequency ECG signals, which are often associated with sharp notches and slurs in the time domain. Furthermore, we set sampling frequency of HFECG signal to 1 kHz so that all the components below 500 Hz could be analyzed according to the sampling theorem. Thereafter, by the virtue of the multifractal theory, we investigated the arithmetic mean and dispersing degree computed from singularity spectrum area of synchronous 12-lead ECG signals obtained from different crowds of human subjects. Variance analysis tests revealed all 12-lead ECG signals from above cohorts were statistically identifiable using these two indicators. The experimental results suggest that the arithmetic mean of the area of the 12-lead ECG signals was apparently large for healthy young but small for myocardial infarction (MI) sufferers. Besides, the dispersing degree of the area of the 12-lead ECG signals was obviously small for healthy young but large for MI sufferers. For the other cohorts, such as ischemia sufferers and healthy elderly, these two values were of middle magnitudes. This indicates that with deeper lesions, the fractal-like structure of the heartbeat system is damaged or structurally changed, which may lead to decline in the nonlinear complexity of the system and concomitant increase in the irregularity and anisotropic propagation of ECG signal. In addition, we found that the 12-lead mean value of singularity spectrum area of human ECG signals can reflect the self-discipline control status of human autonomic nerve to some degree. This value gradually decreased with aging. These findings suggest that the self-discipline control of the human autonomic nervous system weakens with aging. The nonlinear complexity of ECG signal then descends and tends to turn from multifractality to monofractality, implying weakened human individual adaptabilities. Our studies on the nonlinear dynamical features of heart electrical signals and ECG variations with age, disease, and human autonomic nerve control, demonstrate theoretical and diagnostic significances.