Clinical evaluation of the new 12-lead ECG noninvasive epi-endocardial mapping technology

Abstract Funding Acknowledgements Type of funding sources: Private company. Main funding source(s): XSpline S.p.a. Aim Noninvasive electrocardiographic mapping systems may require multichannel (150-250) ECG body surface recordings which is a huge limitation for the adoption into clinical practice. However, most of the currently available noninvasive mapping systems accurately represent epicardial activation. Over the last few years, various technical solutions have been implemented to increase the accuracy of ECG forward-inverse calculations including machine learning algorithms. We developed a novel 12-lead ECG-based, fully automated system for noninvasive full transmural (epi-endocardial) mapping. The aim of this study was to evaluate accuracy of this system in heart failure patients with typical LBBB. Methods Eight consecutive patients (median age: 65; 25-75% range 59-64; 5 male) with typical LBBB QRS morphology (QRS duration 146–224 ms), scheduled for CRT implantation underwent a cardiac CT scan and 12-lead ECG recording, followed by detailed bi-ventricular endocardial electroanatomical contact mapping (EAM). A median of 258 (181-576) local electrograms in LV and 86 (70-139) in RV were collected during EAM. Cardiac CT data were semiautomatically processed using a Unet-like neural network to obtain meshes containing the heart (each chamber with both endocardial and epicardial contour as well as coronary sinus and its branches), lungs and torso using different segmentation module. Noninvasive epi-endocardial activation maps were calculated using state-of-the-art mathematical models including machine learning methods. These data were compared with EAM from the invasive mapping system using Spearman correlation (r), mean absolute error (MAE) and relative distance metrics. Results All early and latest activation patterns were correctly identified (see figure). The average correlation (r) of endocardial activation maps was 0.92 for all cases (min-max: 0.82-0.96). MAE was rather small (9-15 mm) when the non-invasive late activation zone nearly matches the CARTO data (average distance error 14 mm). Overall, the QRS complex was correctly reproduced by our method in all cases (r = 0.96). Conclusions This study shows that new mapping technique can accurately reconstruct endocardial electrical activation maps based on 12-lead ECG and cardiac CT scan. The approach may be very useful in pre-procedural planning of device implantation and catheter ablation. Larger validation studies are however needed for confirming method accuracy at epicardial site.