Abstract:The proposed method re-frames traditional inverse problems of electrocardiography into regression problems, constraining the solution space by decomposing signals with multidimensional Gaussian impulse basis functions. Impulse HSPs were generated with single Gaussian basis functions at discrete heart surface locations and projected to corresponding BSPs using a volume conductor torso model. Both BSP (inputs) and HSP (outputs) were mapped to regular 2D surface meshes and used to train a neural network. Predictive capabilities of the network were tested with unseen synthetic and experimental data. A dense full connected single hidden layer neural network was trained to map body surface impulses to heart surface Gaussian basis functions for reconstructing HSP. Synthetic pulses moving across the heart surface were predicted from the neural network with root mean squared error of $9.1\pm1.4$%. Predicted signals were robust to noise up to 20 dB and errors due to displacement and rotation of the heart within the torso were bounded and predictable. A shift of the heart 40 mm toward the spine resulted in a 4\% increase in signal feature localization error. The set of training impulse function data could be reduced and prediction error remained bounded. Recorded HSPs from in-vitro pig hearts were reliably decomposed using space-time Gaussian basis functions. Predicted HSPs for left-ventricular pacing had a mean absolute error of $10.4\pm11.4$ ms. Other pacing scenarios were analyzed with similar success. Conclusion: Impulses from Gaussian basis functions are potentially an effective and robust way to train simple neural network data models for reconstructing HSPs from decomposed BSPs. The HSPs predicted by the neural network can be used to generate activation maps that non-invasively identify features of cardiac electrical dysfunction and can guide subsequent treatment options.

What problem does this paper attempt to address?

This paper aims to solve the problem of predicting Heart Surface Potentials (HSPs) from Body Surface Potentials (BSPs), that is, the inverse problem of electrocardiogram. Specifically, the author developed, trained and tested a neural network model to predict HSPs from BSPs. This method improves the accuracy and robustness of prediction by reframing the traditional inverse problem as a regression problem and restricting the solution space by decomposing the signal with multi - dimensional Gaussian pulse - basis functions. ### Main contributions of the paper include: 1. **Model and method**: - **Gaussian pulse - basis functions**: The paper shows how to effectively model and generate HSPs using three - dimensional Gaussian basis functions. - **Neural network design**: A simple fully - connected single - hidden - layer neural network architecture is proposed, which can predict unseen physiologically relevant synthetic HSPs from BSPs. - **Experimental verification**: The effectiveness of the method is verified by making predictions on experimentally recorded cardiac data. 2. **Data generation**: - Use a volume conductor model to generate Gaussian pulse - basis function pairs of HSP and BSP in the heart - torso domain. - Gaussian basis functions are generated on a 9×12 heart - surface grid, with parameters including position $(\mu_x, \mu_y)$, width $\sigma$ and amplitude $A$. 3. **Training and testing**: - **Training data**: 129,600 different Gaussian pulses are generated for training and testing the neural network. - **Training process**: Use a 70 - 30 training - validation split, optimize hyperparameters through grid search, and train using the ADAM optimizer and the root - mean - square loss function. 4. **Prediction and evaluation**: - **Moving Gaussian signal**: Generate Gaussian signals moving along the heart surface to test the prediction ability of the neural network. - **Robustness test**: Evaluate the influence of noise, changes in heart position and reduction of the training set on the prediction results. - **Experimental data**: Use experimentally recorded porcine heart data to verify the effectiveness of the method in practical applications. ### Results and conclusions: - **Moving Gaussian signal prediction**: The neural network can reliably reconstruct continuous Gaussian signals moving on the heart surface, with an average RMSE of 9.12±1.37%. - **Robustness**: The prediction results are somewhat robust to noise and changes in heart position, especially with a limited increase in error at a displacement of 40 mm. - **Experimental data prediction**: For the left - ventricular pacing activation time recorded experimentally, the absolute difference between the HSPs predicted by the neural network and the experimental data is 10.4±11.4 ms, and the average RMSE is 16.5±4.18%. ### Significance: - **Non - invasive analysis**: By predicting HSPs, activation maps can be generated to non - invasively identify the characteristics of cardiac electrical dysfunction and guide subsequent treatment options. - **Generality**: By learning in the Gaussian function space, this method can represent electrocardiogram signals under different diseases and drug states and has wide applicability. In general, this paper provides an effective neural network method that can accurately predict HSPs under non - invasive conditions, providing a new tool for cardiac pathology research and clinical applications.

Impulse data models for the inverse problem of electrocardiography

Application of kernel principal component analysis and support vector regression for reconstruction of cardiac transmembrane potentials.

Studies of Localization of the Ventricular Preexcitation Accessory Pathway Using Heart Model Parameter Optimization

Solving the Electrocardiography Inverse Problem by Using an Optimal Algorithm Based on the Total Least Squares Theory

Effect of Cardiac Motion on Solution of the Electrocardiography Inverse Problem.

The Combination of Self-Organizing Feature Maps and Support Vector Regression for Solving the Inverse ECG Problem

Spatial Sparse Constraint in the Transmembrane Potential Based Ecg Inverse Problem

A Comparison of Different Choices for the Regularization Parameter in Inverse Electrocardiography Models

Deep learning neural nets for detecting heart activity

Physiological Model Based Deep Learning Framework for Cardiac TMP Recovery

Physiological-Model-Constrained Noninvasive Reconstruction of Volumetric Myocardial Transmembrane Potentials

Physiology and Geometry-Informed Deep Learning Framework for 12-Lead ECG Cardiac Transmembrane Potential Recovery

A machine learning approach to reconstruction of heart surface potentials from body surface potentials

NONINVASIVE IMAGING OF 3D CARDIAC ELECTROPHYSIOLOGY

12-Lead ECG-Driven Cardiac TMP Recovery VIA a Joint Physiological Model

Cardiac Transmembrane Potential Imaging with GCN Based Iterative Soft Threshold Network

Forward-Solution Aided Deep-Learning Framework for Patient-Specific Noninvasive Cardiac Ectopic Pacing Localization

A Statistical Physiological-Model-Constrained Framework for Computational Imaging of Subject-Specific Volumetric Cardiac Electrophysiology Using Optical Imaging and MRI Data

Imaging of 3D Cardiac Electrical Activity: A Model-Based Recovery Framework

Inference of ventricular activation properties from non-invasive electrocardiography

Beat-by-Beat Estimation of Hemodynamic Parameters in Left Ventricle Based on Phonocardiogram and Photoplethysmography Signals Using a Deep Learning Model: Preliminary Study