An Algorithm for Non-Invasive Mapping Based on Cardiac Anatomy and 12-Lead Electrocardiogram Data

Cardiac mapping provides detailed insights into a patient's unique heart anatomy and electrical patterns. In this paper, we propose an algorithm that integrates cardiac anatomical information with 12-lead electrocardiogram (ECG) data to create activation maps without invasive procedures. To find the activation map of the corresponding ECG of the patient, we solve the multiple forward ECG problems. We use the anisotropic eikonal equation in conjunction with the Lead-Field approach to compute the ECG. By varying the activation area parameters and conductivity parameters in the eikonal equation we generate a subset of ECGs. We use a variational autoencoder to parameterise the 12-channel ECG. Then we train a surrogate model on the generated dataset and apply it to find the closest solutions to the clinical ECG. To calculate the activation map, we use the average solution obtained from the found range of parameters that give close solutions to the patient's ECG. Electroanatomical mapping (EAM) data from the left ventricular endocardium of five patients were used to validate the algorithm. Comparing local activation times, the mean absolute error was 16 ms and a mean correlation of 0.82 was obtained between EAM endocardial activation times and the proposed algorithm.