AI-Based Real-Time ECG Anomaly Detection Using Temporal Convolutional Networks: A Comprehensive Review

Automatic detection and alarm of abnormal electrocardiogram (ECG) events play an important role in an ECG monitor system; however, popular classification models based on supervised learning fail to detect abnormal ECG effectively. Thus, we propose an ECG anomaly detection framework (ECG-AAE) based on an adversarial autoencoder and temporal convolutional network (TCN) which consists of three modules (autoencoder, discriminator, and outlier detector). The ECG-AAE framework is trained only with normal ECG data. Normal ECG signals could be mapped into latent feature space and then reconstructed as the original ECG signal back in our model, while abnormal ECG signals could not. Here, the TCN is employed to extract features of normal ECG data. Then, our model is evaluated on an MIT-BIH arrhythmia dataset and CMUH dataset, with an accuracy, precision, recall, F1-score, and AUC of 0.9673, 0.9854, 0.9486, 0.9666, and 0.9672 and of 0.9358, 0.9816, 0.8882, 0.9325, and 0.9358, respectively. The result indicates that the ECG-AAE can detect abnormal ECG efficiently, with its performance better than other popular outlier detection methods.

Heart disease is still a heavy hitter on the list of leading causes of death around the globe, so getting ahead of the game with early and accurate predictions is crucial for effective treatment. Traditional diagnostic systems often put all their eggs in one basket, relying solely on clinical data or electrocardiogram (ECG) signals. This one-track approach can really throw a wrench in the works, limiting their predictive power. To tackle this head-on, we suggest a Hybrid Deep Learning Framework for Heart Disease Prediction Using Multi-Modal Clinical and ECG Data, which weaves together cutting-edge neural architectures, attention mechanisms, and ensemble strategies to boost performance and clarity.The framework gets the ball rolling by preprocessing UCI Cleveland clinical features with a bit of normalization and encoding, while ECG signals are put through the wringer with noise removal, segmentation, and scaling. Clinical data are shaped using machine learning classifiers and ANN, while ECG features are pulled together with CNN, BiLSTM, and Transformer layers to snag local, temporal, and contextual insights. A Graph Neural Network (GNN) weaves together the threads of clinical and ECG features, while a cross-attention fusion module dances to the rhythm of dynamically aligning and prioritizing the key features from both modalities. Predictions are cooked up using a stacked ensemble classifier, and model interpretability is kept in the clear through SHAP values, attention heatmaps, and graph attention visualization.The experimental evaluation reveals that the proposed method takes the cake, leaving the baseline models in the dust. Our framework hit the nail on the head with a whopping 94.8% accuracy, 93.5% precision, 95.1% recall, and 94.3% F1-score, leaving traditional machine learning in the dust at a mere 86.5%, and unimodal deep learning methods like CNN-BiLSTM and Transformer trailing behind at 89.7% and 90.4%, respectively. Talk about raising the bar!To wrap it all up, the suggested hybrid method hits the nail on the head by blending structured clinical data with ECG signals, providing top-notch predictive accuracy and clarity that’s second to none. This positions it as a strong contender for practical clinical decision support and the early detection of cardiovascular disease. Keywords: Heart Disease Prediction, Hybrid Deep Learning, Multi-Modal Fusion, ECG Signals, Cross-Attention, Graph Neural Network, Explainable AI,

Human physiological data are naturalistic and objective user data inputs for a great number of cyber-physical systems (CPS). Electrocardiogram (ECG) as a widely used physiological golden indicator for certain human state and disease diagnosis is often used as user data input for various CPS such as medical CPS and human–machine interaction. Wireless transmission and wearable technology enable long-term continuous ECG data acquisition for human–CPS interaction; however, these emerging technologies bring challenges of storing and wireless transmitting huge amounts of ECG data, leading to energy efficiency issue of wearable sensors. ECG signal compression technique provides a promising solution for these challenges by decreasing ECG data size. In this study, we develop the first scheme of leveraging empirical mode decomposition (EMD) on ECG signals for sparse feature modeling and compression and further propose a new ECG signal compression framework based on EMD constructed feature dictionary. The proposed method features in compressing ECG signals using a very limited number of feature bases with low computation cost, which significantly improves the compression performance and energy efficiency. Our method is validated with the ECG data from MIT-BIH arrhythmia database and compared with existing methods. The results show that our method achieves the compression ratio (CR) of up to 164 with the root mean square error (RMSE) of 3.48% and the average CR of 88.08 with the RMSE of 5.66%, which is more than twice of the average CR of the state-of-the-art methods with similar recovering error rate of around 5%. For diagnostic distortion perspective, our method achieves high QRS detection performance with the sensitivity (SE) of 99.8% and the specificity (SP) of 99.6%, which shows that our ECG compression method can preserve almost all the QRS features and have no impact on the diagnosis process. In addition, the energy consumption of our method is only 30% of that of other methods when compared under the same recovering error rate.

Any significant alteration in the Electro-Cardio-Gram (ECG) signal wave components (P-QRS-T) for a time duration is detected as arrhythmia. In this paper, a novel fractional wavelet transform (FrWT) is used as a preprocessing tool. FrWT describes the given signal in time–frequency fractional domain using fractional Fourier transform and its denoising using wavelet transform. Because of this novel and intriguing property, it is broadly utilized as a noise removal tool in the fractional domain along with multiresolution analysis. Next, features are extracted using Yule–Walker autoregressive modeling. Dimensionality of the extracted features is to be reduced for proper detection of different types of arrhythmias. Principal component analysis has been applied for arrhythmia detection using variance estimation. The proposed method is evaluated on the basis of various performance parameters such as output SNR, mean squared error (MSE) and detection accuracy ( $$ {\text{DE}}_{\text{Acc}} $$ ). An output SNR of 33.41 dB, MSE of 0.1689% and Acc of 99.94% for real-time ECG database and output SNR of 25.25 dB, MSE of 0.1656%, $$ {\text{DE}}_{\text{Acc}} $$ of 99.89% for MIT-BIH Arrhythmia database are obtained.

Accurate load forecasting is essential for ensuring safe, stable, and economical operation of energy internet. Temporal convolutional networks (TCNs) have demonstrated superior performance, when compared to recurrent neural network models, since their introduction in electrical load forecasting. However, the current TCN‐based models are unable to obtain a large receptive field and strong long‐time feature extraction capability owing to the specific kernel size of the 1D convolution structure. This paper proposes a temporal inception convolutional network based on multi‐head attention (TICN‐Att) for ultra‐short‐term load prediction. By introducing an inception structure into the TCN, the proposed model can extract multi‐dimensional information from the input features, through the multiple hidden convolutional kernels of different scales, without stacking layers depth‐wise. Simultaneously, by introducing a multi‐head attention mechanism, the TICN‐Att model has a long time‐dependent extraction capability, similar to that of the long short‐term memory network models. The generalization and validity of the model are tested using the global energy forecasting competition (GEFCOM2014) dataset, electrical load data of a city in Jiangsu (China), and PJM power system dataset. The experimental results demonstrated that the proposed model has the best prediction effect, compared to the other state‐of‐the‐art models.

Hierarchical deep learning with Generative Adversarial Network for automatic cardiac diagnosis from ECG signals

An explainable attention-based TCN heartbeats classification model for arrhythmia detection

An electrocardiogram (ECG) signal is an important diagnostic tool for cardiologists to detect the abnormality. In continuous monitoring, an ambulatory huge amount of ECG data is involved. This leads to high storage requirements and transmission costs. Hence, to reduce the storage and transmission cost, there is a requirement for an efficient compression or coding technique. One of the most promising compression techniques is Compressive Sensing (CS) which makes efficient compression of signals. By this methodology, a signal can easily be reconstructed if it has a sparse representation. This paper presents the Block Sparse Bayesian Learning (BSBL)-based multiscale compressed sensing (MCS) method for the compression of ECG signals. The main focus of the proposed technique is to achieve a reconstructed signal with less error and more energy efficiency. The ECG signal is sparsely represented by wavelet transform. MIT-BIH Arrhythmia database is used for testing purposes. The Huffman technique is used for encoding and decoding. The signal recovery is appropriate up to 75% of compression. The quality of the signal is ascertained using the standard performance measures such as signal-to-noise ratio (SNR) and Percent root mean square difference (PRD). The quality of the reconstructed ECG signal is also validated through the visual method. This method is most suitable for telemedicine applications.

Abstract. The present work proposes a prediction model of significant wave height (SWH) and average wave period (APD) based on variational mode decomposition (VMD), temporal convolutional networks (TCNs), and long short-term memory (LSTM) networks. The wave sequence features were obtained using VMD technology based on the wave data from the National Data Buoy Center. Then the SWH and APD prediction models were established using TCNs, LSTM, and Bayesian hyperparameter optimization. The VMD–TCN–LSTM model was compared with the VMD–LSTM (without TCN cells) and LSTM (without VMD and TCN cells) models. The VMD–TCN–LSTM model has significant superiority and shows robustness and generality in different buoy prediction experiments. In the 3 h wave forecasts, VMD primarily improved the model performance, while the TCN had less of an influence. In the 12, 24, and 48 h wave forecasts, both VMD and TCNs improved the model performance. The contribution of the TCN to the improvement of the prediction result determination coefficient gradually increased as the forecasting length increased. In the 48 h SWH forecasts, the VMD and TCN improved the determination coefficient by 132.5 % and 36.8 %, respectively. In the 48 h APD forecasts, the VMD and TCN improved the determination coefficient by 119.7 % and 40.9 %, respectively.

Electrocardiogram (ECG) signals are significantly distorted during recording by muscle artifact (MA), causing signal frequency overlap and making it difficult to interpret ECG data correctly. Deep learning (DL) methods for signal processing have shown promising results. However, there is a significant necessity in building proper DL models with appropriate datasets. We propose an enhanced hybrid deep learning framework called HRGB-Net based on residual neural network (ResNet), global channel attention block (GCAB), and bidirectional-long-short-term memory (Bi-LSTM) blocks for filtering the MA noise from ECG by using three distinctive MIT-BIH real-time datasets from the PhysioNet repository by creating suitable datasets for training. We use both raw ECG data and short-time Fourier-transformed (STFT) ECG data for comparative analysis with three neural network models: a convolutional neural Network (CNN), a fully connected neural network (FCNN), and a regression-based LSTM (Reg-LSTM-DNN) model to assess the proposed model. The signal-to-noise ratio (SNR) of noisy ECG signals is varied from - 7dB to 2dB to analyze the mean square error (MSE) and correlation coefficient (CC) performances after the denoising process. Our proposed method utilizes the regression ability to remove MA noise and generate a clean ECG signal with improved values of these signal parameters. The STFT-trained and tested ECG data shows better results than the raw ECG data for efficiently eliminating the MA with a 98.82% correlation coefficient and optimal MSE value of 0.053068. The results prove our proposed HRGB-Net model's remarkable ability to outperform the neural network models and other standard techniques.

Heart rate variability (HRV) analysis is used as a marker of autonomic nervous system activity which may be related to mental and/or physical activity. HRV features can be extracted by detecting QRS complexes from an electrocardiogram (ECG) signal. The difficulties in QRS complex detection are due to the artifacts and noises that may appear in the ECG signal when subjects are performing their daily life activities such as exercise, posture changes, climbing stairs, walking, running, etc. This study describes a strong computation method for real-time QRS complex detection. The detection is improved by the prediction of the position of [Formula: see text] waves by the estimation of the RR intervals lengths. The estimation is done by computing the intensity of the electromyogram noises that appear in the ECG signals and known here in this paper as ECG Trunk Muscles Signals Amplitude (ECG-TMSA). The heart rate (HR) and ECG-TMSA increases with the movement of the subject. We use this property to estimate the lengths of the RR intervals. The method was tested using famous databases, and also with signals acquired when an experiment with 17 subjects from our laboratory. The obtained results using ECG signals from the MIT-Noise Stress Test Database show a QRS complex detection error rate (ER) of 9.06%, a sensitivity of 95.18% and a positive prediction of 95.23%. This method was also tested against MIT-BIH Arrhythmia Database, the result are 99.68% of sensitivity and 99.89% of positive predictivity, with ER of 0.40%. When applied to the signals obtained from the 17 subjects, the algorithm gave an interesting result of 0.00025% as ER, 99.97% as sensitivity and 99.99% as positive predictivity.

Background: Electrocardiograms (ECGs) are frequently stored as printed images, but tools developed for artificial intelligence (AI)-based automated diagnosis and patient phenotyping exclusively rely on signal data, limiting their application. Methods: We developed a model that simultaneously incorporates both signal and image-based ECG data in clinical diagnosis and patient phenotyping. We used all 12-lead ECGs from a large deidentified database from Brazil, with signals transformed to a standard 12-lead ECG image. Both ECG signals and images were randomly subset into training, validation and test sets (90%-5%-5%). We used 6 physician-defined clinical labels spanning rhythm and conduction disorders as labels (atrial fibrillation [AF], sinus tachy/bradycardia [ST/SB], 1 st degree AV block [1dAVB], RBBB, LBBB). In addition, ECG-based identification of patient sex was assessed as a higher dimensional label. Convolutional Neural Networks were constructed with custom Inception net architecture for signals and EfficientNet-B3 for images. Model performance was evaluated in the holdout test set and US-based external validation dataset, PTB-XL. Results: The derivation dataset included 2,228,236 ECGs in 1,506,112 individuals (mean age 54y, 60% female). The ECG-DualNet model successfully labeled all 6 clinical diagnoses and patient gender with a high model performance (AUC of .81-.99 for signal and .92-.99 for images, F1 scores, .31-.78, and .49-82) in the 111,412 held-out test ECGs (Fig A-B). In external validation, with 21,785 ECGs, the model had excellent performance on both ECG signals and images (AUCs .75-.99 and .89-.99, F1 scores, .25-82, and .44-.86) (Fig C-D). Conclusion: A novel multilabel dual signal and image-based model achieved high performance in both clinical diagnosis and higher dimensional phenotypic labels across distinct ECG sources, with the ability to standardize automated diagnosis and patient phenotyping across ECG data formats.

Electrocardiogram (ECG) signal evaluation is routinely used in clinics as a significant diagnostic method for detecting arrhythmia. However, it is very labor intensive to externally evaluate ECG signals, due to their small amplitude. Using automated detection and classification methods in the clinic can assist doctors in making accurate and expeditious diagnoses of diseases. In this study, we developed a classification method for arrhythmia based on the combination of a convolutional neural network and long short-term memory, which was then used to diagnose eight ECG signals, including a normal sinus rhythm. The ECG data of the experiment were derived from the MIT-BIH arrhythmia database. The experimental method mainly consisted of two parts. The input data of the model were two-dimensional grayscale images converted from one-dimensional signals, and detection and classification of the input data was carried out using the combined model. The advantage of this method is that it does not require performing feature extraction or noise filtering on the ECG signal. The experimental results showed that the implemented method demonstrated high classification performance in terms of accuracy, specificity, and sensitivity equal to 99.01%, 99.57%, and 97.67%, respectively. Our proposed model can assist doctors in accurately detecting arrhythmia during routine ECG screening.

Multivariate time-series forecasting is one of the crucial and persistent challenges in time-series forecasting tasks. As a kind of data with multivariate correlation and volatility, multivariate time series impose highly nonlinear time characteristics on the forecasting model. In this paper, a new multivariate time-series forecasting model, multivariate temporal convolutional attention network (MTCAN), based on a self-attentive mechanism is proposed. MTCAN is based on the Convolution Neural Network (CNN) model, using 1D dilated convolution as the basic unit to construct asymmetric blocks, and then, the feature extraction is performed by the self-attention mechanism to finally obtain the prediction results. The input and output lengths of this network can be determined flexibly. The validation of the method is carried out with three different multivariate time-series datasets. The reliability and accuracy of the prediction results are compared with Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Long Short-Term Memory (ConvLSTM), and Temporal Convolutional Network (TCN). The prediction results show that the model proposed in this paper has significantly improved prediction accuracy and generalization.

Research on smartphone-based human activity recognition (HAR) is prevalent in the field of healthcare, especially for elderly activity monitoring. Researchers usually propose to use of accelerometers, gyroscopes or magnetometers that are equipped in smartphones as an individual sensing modality for human activity recognition. However, any of these alone is limited in capturing comprehensive movement information for accurate human activity analysis. Thus, we propose a smartphone-based HAR approach by leveraging the inertial signals captured by these three sensors to classify human activities. These heterogeneous sensors deliver information on various aspects of nature, motion and orientation, offering a richer set of features for more accurate representations of the activities. Hence, a deep learning approach that amalgamates long short-term memory (LSTM) in temporal convolutional network (TCN) is proposed. We use independent temporal convolutional networks, coined as temporal convolutional streams, to independently analyse the temporal data of each sensing modality. We name this architecture multi-stream TC-LSTM. The performance of multi-stream TC-LSTM is assessed on the self-collected elderly activity database. Empirical results exhibit that multi-stream TC-LSTM outperforms the existing machine learning and deep learning models, with an F1 score of 98.3 %