Pre-Processing techniques and artificial intelligence algorithms for electrocardiogram (ECG) signals analysis: A comprehensive review

Effective and powerful methods for denoising real electrocardiogram (ECG) signals are important for wearable sensors and devices. Deep Learning (DL) models have been used extensively in image processing and other domains with great success but only very recently have been used in processing ECG signals. This paper presents several DL models namely Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Restricted Boltzmann Machine (RBM) together with the more conventional filtering methods (low pass filtering, high pass filtering, Notch filtering) and the standard wavelet-based technique for denoising EEG signals. These methods are trained, tested and evaluated on different synthetic and real ECG datasets taken from the MIT PhysioNet database and for different simulation conditions (i.e. various lengths of the ECG signals, single or multiple records). The results show the CNN model is a performant model that can be used for off-line denoising ECG applications where it is satisfactory to train on a clean part of an ECG signal from an ECG record, and then to test on the same ECG signal, which would have some high level of noise added to it. However, for real-time applications or near-real time applications, this task becomes more cumbersome, as the clean part of an ECG signal is very probable to be very limited in size. Therefore the solution put forth in this work is to train a CNN model on 1 second ECG noisy artificial multiple heartbeat data (i.e. ECG at effort), which was generated in a first instance based on few sequences of real signal heartbeat ECG data (i.e. ECG at rest). Afterwards it would be possible to use the trained CNN model in real life situations to denoise the ECG signal.

Enhanced multi-label cardiology diagnosis with channel-wise recurrent fusion

Electrocardiogram (ECG) signal analysis has received special attention of the researchers in the recent past because of its ability to divulge crucial information about the electrophysiology of the heart and the autonomic nervous system activity in a noninvasive manner. Analysis of the ECG signals has been explored using both linear and nonlinear methods. However, the nonlinear methods of ECG signal analysis are gaining popularity because of their robustness in feature extraction and classification. The current study presents a review of the nonlinear signal analysis methods, namely, reconstructed phase space analysis, Lyapunov exponents, correlation dimension, detrended fluctuation analysis (DFA), recurrence plot, Poincaré plot, approximate entropy, and sample entropy along with their recent applications in the ECG signal analysis.

PO-04-175 RECONSTRUCTING TWELVE-LEAD ELECTROCARDIOGRAM SIGNALS FROM A SINGLE-LEAD SIGNAL USING GENERATIVE ADVERSARIAL NETWORKS

Electrocardiogram (ECG) signal shows the different electrical activities of the heart. Since manual analysis is tedious and time-consuming, an offline automated ECG analysis and classification scheme is proposed in this study. Using Discrete Wavelet Transform (DWT) as feature extraction, this comparative study focused on ECG classification between Adaptive Neuro-Fuzzy Inference System (ANFIS) and Support Vector Machine (SVM). It aims to classify normal and abnormal heartbeats with the addition of non-ECG signals. Abnormal heartbeats include ECG signals with atrial fibrillation and ventricular tachycardia. The ECG signals that were used as basis in comparing the results of the two pairings came from MIT-BIH Arrhythmia Database that is found on PhysioNet while the non-ECG signals came from another database which is related to stress recognition. ECG signal analysis in this study comprises three stages: the acquisition of signal from database, the feature extraction and the classification for determining the signal. In this study, SVM using the kernel function Radial Basis Function (RBF) paired with the mother wavelet Daubechies proved to be better than ANFIS paired with Haar mother wavelet.

Real time monitoring and detection of abnormal heart rhythms, specifically through acquisition of electrocardiogram (ECG) signals for extended periods (24 to 48 hours) have been sought by many researchers. Abnormal ECG signals could appear any time in the day during regular patient activities. Spurious noise associated with motion and muscular contraction is usually coupled with the acquired ECG signal. Cancelling muscle contraction and motion artifacts from real time ECG signals is extremely vital for accurate ECG signal classification for analysis by trained medical specialists. The intent of this research was to develop a bi-electrode wireless acquisition system that captured and transmitted ECG signals and motion acceleration information to a host computer system for data capture and analysis. The designed wireless ECG system includes a 3-axis micro electromechanical (MEMS) accelerometer for motion detection.

In this paper, the electrocardiogram (ECG) signal detecting, amplifying, anti-aliasing filtering circuits were designed and made to get the analog ECG signal, and then make it necessarily conditioned to meet the requirement of data acquisition card USB-6008( signal frequency: 0.05Hz ~ 159Hz, signal amplification: + 1000 times). Using the data acquisition card USB-6008 developed by NI Company for A/D conversion, digital form of ECG signal may be gotten, and it can be transmitted to a personal computer for further processing by LabVIEW2009 (evaluation version). We carried out multi-resolution analysis of the digital ECG signal and got ECG signal components in different sub bands. For the purpose of wavelet filtering, the ECG signal was reconstructed with the signal components in sub bands which reflect the characteristics of ECG and ideal effect of ECG was obtained.

The monitoring and processing of electrocardiogram (ECG) beats have been actively studied in recent years: new lines of research have even been developed to analyze ECG signals using mobile devices. Considering these trends, we proposed a simple and low computing cost algorithm to process and analyze an ECG signal. Our approach is based on the use of linear regression to segment the signal, with the goal of detecting the R point of the ECG wave and later, to separate the signal in periods for detecting P, Q, S, and T peaks. After pre-processing of ECG signal to reduce the noise, the algorithm was able to efficiently detect fiducial points, information that is transcendental for diagnosis of heart conditions using machine learning classifiers. When tested on 260 ECG records, the detection approach performed with a Sensitivity of 97.5% for Q-point and 100% for the rest of ECG peaks. Finally, we validated the robustness of our algorithm by developing an ECG sensor to register and transmit the acquired signals to a mobile device in real time.

In the detection of arrhythmia using deep learning, most methods train neural networks to categorize input electrocardiogram (ECG) signals into typical arrhythmia signals using annotated training ECG data. Such methods can neither detect unknown arrhythmias nor explain why the signals are considered as arrhythmic. To detect arrhythmia automatically, this study proposes a method that learns normal ECG signals and can explain the assessment for the signals clearly. Our method builds a model of normal ECG signals using a convolutional neural network and long short-term memory. The model inputs normal ECG signals and performs training to predict the succeeding normal ECG signals. If an abnormal ECG signal is provided, the model predicts the succeeding ECG signal far from the actual ECG signal and can assess whether the input is abnormal. This means that our method can judge any arrhythmia because it requires no prior knowledge regarding the annotations. Experimental results confirm that the proposed method can detect arrhythmias appropriately without learning them. Furthermore, we propose a GUI that can display the location where abnormality is suspected to the user.

Background: Single-lead electrocardiograms (ECGs) are widely available on wearable and portable devices and can be leveraged as screening tools for structural heart disorders (SHDs) with novel deep learning ECG applications. However, scope of model development is limited by the need for large numbers of labeled examples. We present a novel approach to encode unstructured echocardiographic information on 1-lead ECGs to enable the design of wearable-adapted artificial intelligence (AI)-ECG models with efficient development for detecting SHDs. Methods: We linked 274057 unique ECGs from 77378 patients with reports of echocardiograms done within 30 days across a large, diverse health system (2015-2019). We extracted lead I ECG signals that are typically acquired on wearable devices. We defined 2 discrete encoders, a convolutional neural network (CNN) for ECGs and a RoBERTa-based text encoder for echocardiographic reports. We applied contrastive pre-training for 200 epochs to learn a shared signal-text embedding. In a transfer learning experiment involving an age- and sex-matched training pool of temporally-distinct ECGs all from 2019-2023, we trained sequentially smaller amounts to detect LVEF<40%, LVDD (left ventricular diastolic dysfunction), and a composite measure of LVEF<40%, moderate or severe aortic stenosis, aortic regurgitation, mitral regurgitation, and/or severe LVH (IVSd ≥15 mm) using our Wearable-Echo-FM pre-trained versus standard single-lead ECG CNN model. We examined the performance of these models on the same test set. Results: There were a total of 11923 patients (1 ECG per patient, age, 65.3±17.2 2016 (17%) women) in the test set. For models trained on 100% data, the Wearable-Echo-FM model achieved AUROCs comparable to a standard CNN for LVEF < 40% (0.894 vs 0.884), LVDD (0.849 vs 0.843) and a composite of SHDs (AUROC of 0.887 vs 0.869) in the test set. In few-shot learning experiments that only used 0.5% of the training data (~1000 ECGs), the pre-trained model achieved an AUROC of 0.855 vs. 0.548 for the standard model for LVEF < 40%, 0.819 vs. 0.582 for LVDD, and 0.863 vs. 0.496 for the composite SHD. Conclusions: We propose an ECG-Echo foundation model that learns shared embeddings between 1-lead ECGs and complex structural cardiac phenotypes recorded on unstructured echocardiographic reports. Our approach enhances the efficiency of developing AI algorithms for SHD screening on wearable and portable devices.

The current study was designed to study the effect of a patriotic song (stimulus) on the Autonomic Nervous System (ANS) physiology. The alteration of the ANS physiology was estimated from the heart rate variability (HRV) analysis. The electrocardiogram (ECG) signals were acquired in the pre- and post-stimulus conditions. The results suggested an increase in the parasympathetic activity during the post-stimulus condition. The ECG signals were further analyzed in the time domain. Numerous statistical features of the ECG signals were significantly different (with predictor importance > 0.95) during the post-stimulus condition as compared to the pre-stimulus condition thereby suggesting an alteration in the cardiac electrophysiology. The important features, identified from both HRV and ECG signal processing, were classified using radial basis function (RBF) artificial neural network (ANN). A classification efficiency of >85% was obtained.

This paper introduces a new method of simultaneous acquisition of electrocardiogram (ECG) signals and electrical impedance tomography (EIT) measurements on active electrodes, i.e. electrodes that are applying current, in an EIT system. Howland noise reduction followed by an integrate-and-dump filter and adaptive filtering are used to extract the ECG signal without loss of amplitude. Results of simultaneous ECG signals and EIT reconstructions on human subjects are shown. It is also demonstrated that the recovered ECG waveforms from 32 electrodes can be used to reconstruct the heart current source vectors during a cardiac cycle, yielding an approximate solution to the inverse problem of electrocardiography. Acquiring ECG signals from all electrodes attached to the body simultaneously with EIT data acquisition eliminates the need for a separate system to collect ECG signals. Time-aligned ECG signals are helpful in interpreting pulsatile perfusion images by providing exact timing of each image in relation to the cardiac cycle. Additionally, collecting ECG signals from many or all of the EIT electrodes enables reconstruction of the heart's moving dipole moment at the same time as the EIT images.

Recently, a lot of emphasis has been placed on Artificial Intelligence (AI) and Machine Learning (ML) algorithms in medicine and the healthcare industry. Cardiovascular disease (CVD), is one of the most common causes of death globally, and Electrocardiogram (ECG) is the most widely used diagnostic tool to investigate this disease. However, the analysis of ECG signals is a very difficult process. Therefore, in this work, automated classification of ECG data into five different arrhythmia classes is proposed, based on MIT-BIH dataset. Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Deep Learning (DL) models were used. The black-box nature of these complex models imposes the need to explain their outcomes. Hence, both Permutation Feature Importance (PFI) with Gradient-Weighted Class Activation Maps (Grad-CAM) interpretability techniques were investigated. Using the K-Fold cross-validation method, the models achieved an accuracy of 97.1% and 98.5% for CNN and LSTM, respectively.

Smart clothing that can measure electrocardiogram (ECG) signals and monitor the health status of people meets the needs of our increasingly aging society. However, the conventional measurement of ECG signals is complicated and its electrodes can cause irritation to the skin, which makes the conventional measurement method unsuitable for applications in smart clothing. In this paper, a novel wearable measurement of ECG signals is proposed. There are only three ECG textile electrodes knitted into the fabric of smart clothing. The acquired ECG signals can be transmitted to a smartphone via Bluetooth, and they can also be sent out to a PC terminal by a smartphone via WiFi or Internet. To get more significant ECG signals, the ECG differential signal between two electrodes is calculated based on a spherical volume conductor model, and the best positions on the surface of a human body for two textile electrodes to measure ECG signals are simulated by using the body-surface potential mapping (BSPM) data. The results show that position 12 in the lower right and position 11 in the upper left of the human body are the best for the two electrodes to measure ECG signals, and the presented wearable measurement can obtain good performance when one person is under the conditions of sleeping and jogging.

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,