ECG Representation Research Articles

Self-supervised inter–intra period-aware ECG representation learning for detecting atrial fibrillation

Atrial fibrillation is a commonly encountered clinical arrhythmia associated with stroke and increased mortality. Since professional medical knowledge is required for annotation, exploiting a large corpus of ECGs to develop accurate supervised learning-based atrial fibrillation algorithms remains challenging. Self-supervised learning (SSL) is a promising recipe for generalized ECG representation learning, eliminating the dependence on expensive labeling. However, without well-designed incorporations of knowledge related to atrial fibrillation, existing SSL approaches typically suffer from unsatisfactory capture of robust ECG representations. In this paper, we propose an inter–intra period-aware ECG representation learning approach. Considering ECGs of atrial fibrillation patients exhibit the irregularity in RR intervals and the absence of P-waves, we develop specific pre-training tasks for interperiod and intraperiod representations, aiming to learn the single-period stable morphology representation while retaining crucial interperiod features. After further fine-tuning, our approach demonstrates remarkable AUC performances on the BTCH dataset, i.e., 0.953/0.996 for paroxysmal/persistent atrial fibrillation detection. On commonly used benchmarks of CinC2017 and CPSC2021, the generalization capability and effectiveness of our methodology are substantiated with competitive results.

Masked self-supervised ECG representation learning via multiview information bottleneck

Masked self-supervised ECG representation learning via multiview information bottleneck

Graph Enhanced Low-Resource ECG Representation Learning for Emotion Recognition Based on Wearable Internet of Things

Graph Enhanced Low-Resource ECG Representation Learning for Emotion Recognition Based on Wearable Internet of Things

Abstract 13832: XplainScar: Explainable Artificial Intelligence to Identify and Localize Left Ventricular Scar in Hypertrophic Cardiomyopathy (HCM) Using 12-lead Electrocardiogram

Background/Rationale: Myocardial scar, identified by late gadolinium enhancement (LGE) on MRI, is associated with sudden death in HCM. Unlike ECG, MRI is expensive and adversely affected by artifacts from implanted devices. However, little is known about ECG features of LV-scar in HCM. Objective: Develop an ECG-based explainable machine learning method to identify and localize LV-scar in HCM. Method: We retrospectively studied 500 HCM patients (JH HCM Registry) for model development, and 248 patients (UCSF HCM Registry) for validation. All patients underwent MRI and ECHO within 1 year of ECG. LV-LGE (scar) was assessed using QMass. After excluding RV-insertion-point-LGE, the LV was divided into basal, mid, and apical regions for scar detection. Resting 12-lead ECGs were segmented, features were extracted and adjusted for LV-mass, age, sex. We utilized unsupervised and self-supervised ECG representation learning, where patients are partitioned into groups of several sub-clusters, each sharing similar ECG patterns, but with high separation between scar and no-scar classes. In each group, a self-supervised neural net and a fully connected neural net successfully predicted LV-scar (see Figure) and revealed ECG features of scar. Results: Our method identifies LV-scar in the JH-dataset with high precision (90%), sensitivity (95%), specificity (80%), F1-score (90%), and generalizes well to UCSF-data (precision:88%, sensitivity:90%, specificity:78%, F1-score:89%). The top ECG features for basal-scar are Q-amplitude, Q-slope, non-terminal QRS duration in aVR, and area under QRS and T wave energy in V1-V2. T-wave inversion in V4-V6, area under QRS in V3, TP slope in V3-V4 predicted apical scar. Features selected for mid scar prediction combine features for basal and apical scar. Conclusion: This is the first ECG-based ML model to identify/localize LV-scar in HCM. Our model demonstrates good performance and reveals ECG features of scar in HCM.

Spatiotemporal representation of cardiac vectorcardiogram (VCG) signals

BackgroundVectorcardiogram (VCG) signals monitor both spatial and temporal cardiac electrical activities along three orthogonal planes of the body. However, the absence of spatiotemporal resolution in conventional VCG representations is a major impediment for medical interpretation and clinical usage of VCG. This is especially so because time-domain features of 12-lead ECG, instead of both spatial and temporal characteristics of VCG, are widely used for the automatic assessment of cardiac pathological patterns.Materials and methodsWe present a novel representation approach that captures critical spatiotemporal heart dynamics by displaying the real time motion of VCG cardiac vectors in a 3D space. Such a dynamic display can also be realized with only one lead ECG signal (e.g., ambulatory ECG) through an alternative lag-reconstructed ECG representation from nonlinear dynamics principles. Furthermore, the trajectories are color coded with additional dynamical properties of space-time VCG signals, e.g., the curvature, speed, octant and phase angles to enhance the information visibility.ResultsIn this investigation, spatiotemporal VCG signal representation is used to characterize various spatiotemporal pathological patterns for healthy control (HC), myocardial infarction (MI), atrial fibrillation (AF) and bundle branch block (BBB). The proposed color coding scheme revealed that the spatial locations of the peak of T waves are in the Octant 6 for the majority (i.e., 74 out of 80) of healthy recordings in the PhysioNet PTB database. In contrast, the peak of T waves from 31.79% (117/368) of MI subjects are found to remain in Octant 6 and the rest (68.21%) spread over all other octants. The spatiotemporal VCG signal representation is shown to capture the same important heart characteristics as the 12-lead ECG plots and more.ConclusionsSpatiotemporal VCG signal representation is shown to facilitate the characterization of space-time cardiac pathological patterns and enhance the automatic assessment of cardiovascular diseases.

Heart Rate Wireless Monitoring System “Cardiobeat”, Novel Approach to ECG Processing and Representation

The paper concerns the development of the heart rate wearable wireless monitoring system “CardioBeat”. This system has two types of transmission mode, which are total signal transmission and heart rate transmission modes. The developed system uses wireless PAN (Personal Area Network), which operates in low-power mode. It is composed of two main parts: ECG registering circuit with wireless communication module and base station with terminal. The registering circuit has three surface electrodes, ECG amplifier and microcontroller. In total signal transmission mode, it can send data with transmission speed corresponding up to 300 ECG samples per second. In heart rate transmission mode, it can calculate heart rate from ECG data with 300 samples per second and send packets at the specified rate. Base station communication unit forwards the received data to PC, where the data can be stored and displayed. The developed equipment showed the possibility of real-time monitoring of the patients in their daily life and is expected to contribute, in particular, to the saving of medical expenses.

Keep the QT interval: It is a reliable predictor of ventricular arrhythmias

The cardiovascular community understands that the QT interval is the surface electrocardiographic representation of ventricular repolarization. However, despite tremendous advances in our understanding of the molecular and cellular basis for cardiac repolarization and its surface ECG representation, clinicians, regulators, and drug developers are increasingly faced with uncertainty over the best way to interpret this parameter. The evidence described below demonstrates that Prolongation of the QT interval is a reliable predictor of ventricular arrhythmias in some settings, and may be less predictable in others The arrhythmogenic potential of a given degree of QT prolongation varies among individuals and drugs. Further, the relationship between QT prolongation and arrhythmia susceptibility is non-linear. Multiple mechanisms may underlie QT prolongation, and these may carry different arrhythmogenic potential. There is little doubt that further understanding of the cellular and molecular mechanisms underlying repolarization hold the promise of developing better indices of the arrhythmogenic potential of abnormalities of repolarization, but that science has not yet matured

Doppler echocardiography and myocardial dyssynchrony: a practical update of old and new ultrasound technologies

Morbidity and mortality rates are higher in patients with severe left ventricular (LV) systolic dysfunction and ECG-derived prolonged QRS interval than in those with normal QRS duration. QRS duration is currently used on the grounds that it reflects the presence of ventricular dyssynchrony. However, 30–40% of patients selected on the basis of a prolonged QRS do not receive benefit by cardiac resynchronization therapy (CRT) since they do not show any significant inverse LV remodeling and QRS duration does not accurately distinguish responders to CRT. Consequently, mechanical dyssynchrony (particularly intra-ventricular dyssynchrony) seems to be much more important than electrical dyssinchrony. Pre- and post-echocardiographic assessment should require the combination of conventional and specific applications ranging from M-mode and pulsed/continuous Doppler, to pulsed Tissue Doppler, the off-line analysis of colour Tissue Velocity Imaging, Strain Rate Imaging, and real time three-dimensional reconstruction However, there is not no consensus about the best approach and the best ultrasound parameter for selecting candidates to CRT and ECG representation of abnormal cardiac conduction still remains as the main criterion in guidelines. This review is a practical update of ultrasound methods and measurements of atrio-ventricular, inter-ventricular and intra-ventricular dyssynchrony and describes experiences which used either conventional Doppler echocardiography and more advanced techniques. By these experiences, the global amount of LV dyssynchrony seems to be critical: the greater intra-ventricular dyssynchrony, the higher the possibility of significant LV inverse remodeling. After CRT, it is necessary also to evaluate the optimal atrio-ventricular delay and ventricular-ventricular delay setting that maximizes LV systolic function.

ECG waveform feature extraction and its application to automated prognosis

Automated correlation of ECG history for early detection of heart disease, especially among the young, has been a matter of increasing interest. However, each electrocardiogram, recorded say a few months apart, generates anywhere from 600 to 2400 digitized data, so that statistical methods cannot directly be applied. An information compression step suitable for such data is presented in this paper and a prediction procedure is developed for forecasting the waveform changes. Specifically, each ECG lead is digitized and represented by itsz-domain modes. These modes are found to exhibit continuity in time, from month to month and year to year, except in the event of major physiological changes such as after surgery, thus lending themselves ideally to statistic al prediction. To enhance discrimination of the subtle changes inP, QRS, andT complexes, the derivatives of the waves are employed for extraction of the modes. This signifies a departure from previous efforts in ECG representation. Indeed, otherwise, important changes in the waves can remain undetected through mode extraction while the human eye can perceive them rather easily from the recorded traces.