Reconstruction of 12-Lead Electrocardiogram from a Three-Lead Patch-Type Device Using a LSTM Network

A wavelet subband based LSTM model for 12-lead ECG synthesis from reduced lead set

Synthesis of a 12-lead electrocardiogram from a reduced lead set has previously been extensively studied in order to meet patient comfort, minimise complexity, and enable telemonitoring. Traditional methods relied solely on the inter-lead correlation between the standard twelve leads for learning the models. The 12-lead ECG possesses not only inter-lead correlation but also intra-lead correlation. Learning a model that can exploit this spatio-temporal information in the ECG could generate lead signals while preserving important diagnostic information. The proposed approach takes leverage of the enhanced inter-lead correlation of the ECG signal in the wavelet domain. Long-short-term memory (LSTM) networks, which have emerged as a powerful tool for sequential data mining, are a type of recurrent neural network architecture with an inherent capability to capture the spatiotemporal information of the heart signal. This work proposes the deep learning architecture that utilizes the discrete wavelet transform and the LSTM to reconstruct a generic 12-lead ECG from a reduced lead set. The experimental results are evaluated using different diagnostic measures and similarity metrics. The proposed framework is well founded, and accurate reconstruction is possible as it can capture clinically significant features and provides a robust solution against noise.

AI-enhanced reconstruction of the 12-lead electrocardiogram via 3-leads with accurate clinical assessment

The 12-lead electrocardiogram (ECG) is an integral component to the diagnosis of a multitude of cardiovascular conditions. It is performed using a complex set of skin surface electrodes, limiting its use outside traditional clinical settings. We developed an artificial intelligence algorithm, trained over 600,000 clinically acquired ECGs, to explore whether fewer leads as input are sufficient to reconstruct a 12-lead ECG. Two limb leads (I and II) and one precordial lead (V3) were required to generate a reconstructed 12-lead ECG highly correlated with the original ECG. An automatic algorithm for detection of ECG features consistent with acute myocardial infarction (MI) performed similarly for original and reconstructed ECGs (AUC = 0.95). When interpreted by cardiologists, reconstructed ECGs achieved an accuracy of 81.4 ± 5.0% in identifying ECG features of ST-segment elevation MI, comparable with the original 12-lead ECGs (accuracy 84.6 ± 4.6%). These results will impact development efforts to innovate ECG acquisition methods with simplified tools in non-specialized settings.

12-Lead ECG Reconstruction Based on Data From the First Limb Lead

PURPOSE

Electrocardiogram (ECG) data obtained from 12 leads are the most common and informative source for analyzing the cardiovascular system's (CVS) condition in medical practice. However, the large number of electrodes, specific placements on the body, and the need for specialized equipment make the ECG acquisition procedure complex and cumbersome. This raises the challenge of reducing the number of ECG leads by reconstructing missing leads based on available data.

METHODS

Most existing methods for reconstructing missing ECG leads rely on utilizing signals simultaneously from multiple known leads. This study proposes a method for reconstructing ECG data in 12 leads using signal data from the first lead, lead I. Such an approach can significantly simplify the ECG registration procedure. The study demonstrates the effectiveness of using unique models with a developed architecture of artificial neural networks (ANNs) to generate the reconstructed ECG signals. Fragments of ECG from lead I, with a duration of 128 samples and a sampling frequency of 100 Hz, are input to the models. ECG fragments can be extracted from the original signal at arbitrary time points. Each model generates an ECG signal of the same length at its output for the corresponding lead.

RESULTS

Despite existing limitations, the proposed method surpasses known solutions regarding ECG generation quality when using a single lead. The study shows that introducing an additional feature of the direction of the electrical axis of the heart (EAH) as input to the ANN models enhances the generation quality. The quality of ECG generation by the proposed ANN models is found to be dependent on the presence of CVS diseases.

CONCLUSIONS

The developed ECG reconstruction method holds significant potential for use in portable registration devices, screening procedures, and providing support for medical decision-making by healthcare specialists.

Feasibility and validity of using deep learning to reconstruct 12-lead ECG from three‑lead signals

BACKGROUND

In the field of mobile health, portable dynamic electrocardiogram (ECG) monitoring devices often have a limited number of lead electrodes due to considerations, such as portability and battery life. This situation leads to a contradiction between the demand for standard 12‑lead ECG information and the limited number of leads collected by portable devices.

METHODS

This study introduces a composite ECG vector reconstruction network architecture based on convolutional neural network (CNN) combined with recurrent neural network by using leads I, II, and V2. This network is designed to reconstruct three‑lead ECG signals into 12‑lead ECG signals. A 1D CNN abstracts and extracts features from the spatial domain of the ECG signals, and a bidirectional long short-term memory network analyzes the temporal trends in the signals. Then, the ECG signals are inputted into the model in a multilead, single-channel manner.

RESULTS

Under inter-patient conditions, the mean reconstructed Root mean squared error (RMSE) for precordial leads V1, V3, V4, V5, and V6 were 28.7, 17.3, 24.2, 36.5, and 25.5 μV, respectively. The mean overall RMSE and reconstructed Correlation coefficient (CC) were 26.44 μV and 0.9562, respectively.

CONCLUSION

This paper presents a solution and innovative approach for recovering 12‑lead ECG information when only three‑lead information is available. After supplementing with comprehensive leads, we can analyze the cardiac health status more comprehensively across 12 dimensions.

A transferable in-silico augmented ischemic model for virtual myocardial perfusion imaging and myocardial infarction detection

This paper proposes an innovative approach to generate a generalized myocardial ischemia database by modeling the virtual electrophysiology of the heart and the 12-lead electrocardiography projected by the in-silico model can serve as a ready-to-use database for automatic myocardial infarction/ischemia (MI) localization and classification. Although the virtual heart can be created by an established technique combining the cell model with personalized heart geometry to observe the spatial propagation of depolarization and repolarization waves, we developed a strategy based on the clinical pathophysiology of MI to generate a heterogeneous database with a generic heart while maintaining clinical relevance and reduced computational complexity. First, the virtual heart is simplified into 11 regions that match the types and locations, which can be diagnosed by 12-lead ECG; the major arteries were divided into 3-5 segments from the upstream to the downstream based on the general anatomy. Second, the stenosis or infarction of the major or minor coronary artery branches can cause different perfusion drops and infarct sizes. We simulated the ischemic sites in different branches of the arteries by meandering the infarction location to elaborate on possible ECG representations, which alters the infraction's size and changes the transmembrane potential (TMP) of the myocytes associated with different levels of perfusion drop. A total of 8190 different case combinations of cardiac potentials with ischemia and MI were simulated, and the corresponding ECGs were generated by forward calculations. Finally, we trained and validated our in-silico database with a sparse representation classification (SRC) and tested the transferability of the model on the real-world Physikalisch Technische Bundesanstalt (PTB) database. The overall accuracies for localizing the MI region on the PTB data achieved 0.86, which is only 2% drop compared to that derived from the simulated database (0.88). In summary, we have shown a proof-of-concept for transferring an in-silico model to real-world database to compensate for insufficient data.

AI-Enhanced Reconstruction of the 12-Lead Electrocardiogram via 3-Leads with Accurate Clinical Assessment

The 12-lead electrocardiogram (ECG) is an integral component to the diagnosis of a multitude of cardiovascular conditions. It is performed using a complex set of skin surface electrodes, limiting its use outside traditional clinical settings. We developed an artificial intelligence algorithm, trained over 600,000 clinically acquired ECGs, to explore whether fewer leads as input are sufficient to reconstruct a full 12-lead ECG. Two limb leads (I and II) and one precordial lead (V3) were required to generate a reconstructed synthetic 12-lead ECG highly correlated with the original ECG. An automatic algorithm for detection of acute myocardial infarction (MI) performed similarly for original and reconstructed ECGs (AUC=0.94). When interpreted by cardiologists, reconstructed ECGs achieved an accuracy of 81.4±5.0% in identifying ST elevation MI, comparable with the original 12-lead ECGs (accuracy 84.6±4.6%). These results will impact development efforts to innovate ECG acquisition methods with simplified tools in non-specialized settings.

Classification feasibility test on multi-lead electrocardiography signals generated from single-lead electrocardiography signals

Nowadays, Electrocardiogram (ECG) signals can be measured using wearable devices, such as smart watches. Most wearable devices provide only a few details; however, they have the advantage of recording data in real time. In this study, 12-lead ECG signals were generated from lead I and their feasibility was tested to obtain more details. The 12-lead ECG signals were generated using a U-net-based generative adversarial network (GAN) that was trained on ECG data obtained from the Asan Medical Center. Subsequently, unseen PTB-XL PhysioNet data were used to produce real 12-lead ECG signals for classification. The generated and real 12-lead ECG signals were then compared using a ResNet classification model; and the normal, atrial fibrillation (A-fib), left bundle branch block (LBBB), right bundle branch block (RBBB), left ventricular hypertrophy (LVH), and right ventricular hypertrophy (RVH) were classified. The mean precision, recall, and f1-score for the real 12-lead ECG signals are 0.70, 0.72, and 0.70, and that for the generated 12-lead ECG signals are 0.82, 0.80, and 0.81, respectively. In our study, according to the result generated 12-lead ECG signals performed better than real 12-lead ECG.

Design and technical validation to generate a synthetic 12-lead electrocardiogram dataset to promote artificial intelligence research

Purpose

The purpose of this study is to construct a synthetic dataset of ECG signal that overcomes the sensitivity of personal information and the complexity of disclosure policies.

Methods

The public dataset was constructed by generating synthetic data based on the deep learning model using a convolution neural network (CNN) and bi-directional long short-term memory (Bi-LSTM), and the effectiveness of the dataset was verified by developing classification models for ECG diagnoses.

Results

The synthetic 12-lead ECG dataset generated consists of a total of 6000 ECGs, with normal and 5 abnormal groups. The synthetic ECG signal has a waveform pattern similar to the original ECG signal, the average RMSE between the two signals is 0.042 µV, and the average cosine similarity is 0.993. In addition, five classification models were developed to verify the effect of the synthetic dataset and showed performance similar to that of the model made with the actual dataset. In particular, even when the real dataset was applied as a test set to the classification model trained with the synthetic dataset, the classification performance of all models showed high accuracy (average accuracy 93.41%).

Conclusion

The synthetic 12-lead ECG dataset was confirmed to perform similarly to the real-world 12-lead ECG in the classification model. This implies that a synthetic dataset can perform similarly to a real dataset in clinical research using AI. The synthetic dataset generation process in this study provides a way to overcome the medical data disclosure challenges constrained by privacy rights, a way to encourage open data policies, and contribute significantly to promoting cardiovascular disease research.

Standardized Database of 12-Lead Electrocardiograms with a Common Standard for the Promotion of Cardiovascular Research: KURIAS-ECG

OBJECTIVES

Electrocardiography (ECG)-based diagnosis by experts cannot maintain uniform quality because individual differences may occur. Previous public databases can be used for clinical studies, but there is no common standard that would allow databases to be combined. For this reason, it is difficult to conduct research that derives results by combining databases. Recent commercial ECG machines offer diagnoses similar to those of a physician. Therefore, the purpose of this study was to construct a standardized ECG database using computerized diagnoses.

METHODS

The constructed database was standardized using Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) and Observational Medical Outcomes Partnership-common data model (OMOP-CDM), and data were then categorized into 10 groups based on the Minnesota classification. In addition, to extract high-quality waveforms, poor-quality ECGs were removed, and database bias was minimized by extracting at least 2,000 cases for each group. To check database quality, the difference in baseline displacement according to whether poor ECGs were removed was analyzed, and the usefulness of the database was verified with seven classification models using waveforms.

RESULTS

The standardized KURIAS-ECG database consists of high-quality ECGs from 13,862 patients, with about 20,000 data points, making it possible to obtain more than 2,000 for each Minnesota classification. An artificial intelligence classification model using the data extracted through SNOMED-CT showed an average accuracy of 88.03%.

CONCLUSIONS

The KURIAS-ECG database contains standardized ECG data extracted from various machines. The proposed protocol should promote cardiovascular disease research using big data and artificial intelligence.

Personalized LSTM Models for ECG Lead Transformations Led to Fewer Diagnostic Errors Than Generalized Models: Deriving 12-Lead ECG from Lead II, V2, and V6

BACKGROUND AND OBJECTIVE

The prevalence of chronic cardiovascular diseases (CVDs) has risen globally, nearly doubling from 1990 to 2019. ECG is a simple, non-invasive measurement that can help identify CVDs at an early and treatable stage. A multi-lead ECG, up to 15 leads in a wearable form factor, is desirable. We seek to derive multiple ECG leads from a select subset of leads so that the number of electrodes can be reduced in line with a patient-friendly wearable device. We further compare personalized derivations to generalized derivations.

METHODS

Long-Short Term Memory (LSTM) networks using Lead II, V2, and V6 as input are trained to obtain generalized models using Bayesian Optimization for hyperparameter tuning for all patients and personalized models for each patient by applying transfer learning to the generalized models. We compare quantitatively using error metrics Root Mean Square Error (RMSE), R², and Pearson correlation (ρ). We compare qualitatively by matching ECG interpretations of board-certified cardiologists.

RESULTS

ECG interpretations from personalized models, when corrected for an intra-observer variance, were identical to the original ECGs, whereas generalized models led to errors. Mean performance values for generalized and personalized models were (RMSE-74.31 µV, R²-72.05, ρ-0.88) and (RMSE-26.27 µV, R²-96.38, ρ-0.98), respectively.

CONCLUSIONS

Diagnostic accuracy based on derived ECG is the most critical validation of ECG derivation methods. Personalized transformation should be sought to derive ECGs. Performing a personalized calibration step to wearable ECG systems and LSTM networks could yield ambulatory 15-lead ECGs with accuracy comparable to clinical ECGs.

Electrocardiogram lead conversion from single-lead blindly-segmented signals

BACKGROUND

The standard configuration's set of twelve electrocardiogram (ECG) leads is optimal for the medical diagnosis of diverse cardiac conditions. However, it requires ten electrodes on the patient's limbs and chest, which is uncomfortable and cumbersome. Interlead conversion methods can reconstruct missing leads and enable more comfortable acquisitions, including in wearable devices, while still allowing for adequate diagnoses. Currently, methodologies for interlead ECG conversion either require multiple reference (input) leads and/or require input signals to be temporally aligned considering the ECG landmarks.

METHODS

Unlike the methods in the literature, this paper studies the possibility of converting ECG signals into all twelve standard configuration leads using signal segments from only one reference lead, without temporal alignment (blindly-segmented). The proposed methodology is based on a deep learning encoder-decoder U-Net architecture, which is compared with adaptations based on convolutional autoencoders and label refinement networks. Moreover, the method is explored for conversion with one single shared encoder or multiple individual encoders for each lead.

RESULTS

Despite the more challenging settings, the proposed methodology was able to attain state-of-the-art level performance in multiple target leads, and both lead I and lead II seem especially suitable to convert certain sets of leads. In cross-database tests, the methodology offered promising results despite acquisition setup differences. Furthermore, results show that the presence of medical conditions does not have a considerable effect on the method's performance.

CONCLUSIONS

This study shows the feasibility of converting ECG signals using single-lead blindly-segmented inputs. Although the results are promising, further efforts should be devoted towards the improvement of the methodologies, especially the robustness to diverse acquisition setups, in order to be applicable to cardiac health monitoring in wearable devices and less obtrusive clinical scenarios.

GRU Neural Network Improved Bioimpedance Based Stroke Volume Estimation during Ergometry Stress Test

Cardiovascular diseases (CVDs) are one of the leading members of non-communicable diseases. An early diagnosis is essential for effective treatment, to reduce hospitalization time and health care costs. Nowadays, an exercise stress test on an ergometer is used to identify CVDs. To improve the accuracy of diagnostics, the hemodynamic status and parameters of a person can be investigated. For hemodynamic management, thoracic electrical bioimpedance has recently been used. This technique offers beat-to-beat stroke volume calculation but suffers from an artifact-sensitive signal that makes such measurements difficult during movement. We propose a new method based on a gated recurrent unit (GRU) neural network and the ECG signal to improve the measurement of bioimpedance signals, reduce artifacts and calculate hemodynamic parameters. We conducted a study with 23 subjects. The new approach is compared to ensemble averaging, scaled Fourier linear combiner, adaptive filter, and simple neural networks. The GRU neural network performs better with single artifact events than shallow neural networks (mean error -0.0244, mean square error 0.0181 for normalized stroke volume). The GRU network is superior to other algorithms using time-correlated data for the exercise stress test.

Smart ECG Biosensor Design with an Improved ANN Performance Based on the Taguchi Optimizer

This paper aims to design a smart biosensor to predict electrocardiogram (ECG) signals in a specific auscultation site from other ECG signals measured from other measurement sites. The proposed design is based on a hybrid architecture using the Artificial Neural Networks (ANNs) model and Taguchi optimizer to avoid the ANN issues related to hyperparameters and to improve its accuracy. The proposed approach aims to optimize the number and type of inputs to be considered for the ANN model. Indeed, different combinations are considered in order to find the optimal input combination for the best prediction quality. By identifying the factors that influence a model’s prediction and their degree of importance via the modified Taguchi optimizer, the developed biosensor improves the prediction accuracy of ECG signals collected from different auscultation sites compared to the ANN-based biosensor. Based on an actual database, the simulation results show that this improvement is significant; it can reach more than 94% accuracy.

Label noise and self-learning label correction in cardiac abnormalities classification

OBJECTIVE

Learning to classify cardiac abnormalities requires large and high-quality labeled datasets, which is a challenge in medical applications. Small datasets from various sources are often aggregated to meet this requirement, resulting in a final dataset prone to label noise owing to inter- and intra-observer variability, and different expertise. It is well known that label noise can affect the performance and generalizability of the trained models. In this work, we explore the impact of label noise and self-learning label correction on the classification of cardiac abnormalities on large heterogeneous datasets of electrocardiogram (ECG) signals.

APPROACH

A state-of-the-art self-learning multi-class label correction method for image classification is adapted to learn a multi-label classifier for electrocardiogram signals. We evaluated our performance using 5-fold cross-validation on the publicly available PhysioNet/Computing in Cardiology (CinC) 2021 Challenge data, with full and reduced sets of leads. Due to the unknown label noise in the testing set, we tested our approach on the MNIST dataset. We investigated the performance under different levels of structured label noise for both datasets.

MAIN RESULTS

Under high levels of noise, the cross-validation results of self-learning label correction showed an improvement of approximately 3% in the Challenge score for the PhysioNet/CinC 2021 Challenge dataset and, an improvement in accuracy of 5$\%$ and reduction of the expected calibration error of 0.03 for the MNIST dataset. We demonstrate that self-learning label correction can be used to effectively deal with the presence of unknown label noise, also when using a reduced number of ECG leads.

Deep-Learning-Based Estimation of the Spatial QRS-T Angle from Reduced-Lead ECGs

The spatial QRS-T angle is a promising health indicator for risk stratification of sudden cardiac death (SCD). Thus far, the angle is estimated solely from 12-lead electrocardiogram (ECG) systems uncomfortable for ambulatory monitoring. Methods to estimate QRS-T angles from reduced-lead ECGs registered with consumer healthcare devices would, therefore, facilitate ambulatory monitoring. (1) Objective: Develop a method to estimate spatial QRS-T angles from reduced-lead ECGs. (2) Approach: We designed a deep learning model to locate the QRS and T wave vectors necessary for computing the QRS-T angle. We implemented an original loss function to guide the model in the 3D space to search for each vector's coordinates. A gradual reduction of ECG leads from the largest publicly available dataset of clinical 12-lead ECG recordings (PTB-XL) is used for training and validation. (3) Results: The spatial QRS-T angle can be estimated from leads {I, II, aVF, V2} with sufficient accuracy (absolute mean and median errors of 11.4° and 7.3°) for detecting abnormal angles without sacrificing patient comfortability. (4) Significance: Our model could enable ambulatory monitoring of spatial QRS-T angles using patch- or textile-based ECG devices. Populations at risk of SCD, like chronic cardiac and kidney disease patients, might benefit from this technology.

Synthesis of standard 12‑lead electrocardiograms using two-dimensional generative adversarial networks

This paper proposes a two-dimensional (2D) bidirectional long short-term memory generative adversarial network (GAN) to produce synthetic standard 12-lead ECGs corresponding to four types of signals-left ventricular hypertrophy (LVH), left branch bundle block (LBBB), acute myocardial infarction (ACUTMI), and Normal. It uses a fully automatic end-to-end process to generate and verify the synthetic ECGs that does not require any visual inspection. The proposed model is able to produce synthetic standard 12-lead ECG signals with success rates of 98% for LVH, 93% for LBBB, 79% for ACUTMI, and 59% for Normal. Statistical evaluation of the data confirms that the synthetic ECGs are not biased towards or overfitted to the training ECGs, and span a wide range of morphological features. This study demonstrates that it is feasible to use a 2D GAN to produce standard 12-lead ECGs suitable to augment artificially a diverse database of real ECGs, thus providing a possible solution to the demand for extensive ECG datasets.

Lead Reconstruction Using Artificial Neural Networks for Ambulatory ECG Acquisition

One of the most powerful techniques to diagnose cardiovascular diseases is to analyze the electrocardiogram (ECG). To increase diagnostic sensitivity, the ECG might need to be acquired using an ambulatory system, as symptoms may occur during a patient’s daily life. In this paper, we propose using an ambulatory ECG (aECG) recording device with a low number of leads and then estimating the views that would have been obtained with a standard ECG location, reconstructing the complete Standard 12-Lead System, the most widely used system for diagnosis by cardiologists. Four approaches have been explored, including Linear Regression with ECG segmentation and Artificial Neural Networks (ANN). The best reconstruction algorithm is based on ANN, which reconstructs the actual ECG signal with high precision, as the results bring a high accuracy (RMS Error < 13 μV and CC > 99.7%) for the set of patients analyzed in this paper. This study supports the hypothesis that it is possible to reconstruct the Standard 12-Lead System using an aECG recording device with less leads.

Automatic Classification of Myocardial Infarction Using Spline Representation of Single-Lead Derived Vectorcardiography

Myocardial infarction (MI) is one of the most prevalent cardiovascular diseases worldwide and most patients suffer from MI without awareness. Therefore, early diagnosis and timely treatment are crucial to guarantee the life safety of MI patients. Most wearable monitoring devices only provide single-lead electrocardiography (ECG), which represents a major limitation for their applicability in diagnosis of MI. Incorporating the derived vectorcardiography (VCG) techniques can help monitor the three-dimensional electrical activities of human hearts. This study presents a patient-specific reconstruction method based on long short-term memory (LSTM) network to exploit both intra- and inter-lead correlations of ECG signals. MI-induced changes in the morphological and temporal wave features are extracted from the derived VCG using spline approximation. After the feature extraction, a classifier based on multilayer perceptron network is used for MI classification. Experiments on PTB diagnostic database demonstrate that the proposed system achieved satisfactory performance to differentiating MI patients from healthy subjects and to localizing the infarcted area.