Background. Noninvasive cardiac imaging (ECGI) demonstrates acceptable accuracy in solving certain electrophysiological problems (for example, identifying the localization of ectopic foci of premature beats) while showing low accuracy in others, such as determining the myocardial activation wavefront.
Aim. To clarify the reasons for this discrepancy and to propose a method for feasibility analysis of noninvasive cardiac mapping.
Methods. Three types of ventricular excitation patterns—premature ventricular contractions, myocardial fibrosis, and ventricular fibrillation—were simulated using six personalized CT derived heart–torso models and the bidomain model with the TNNP 2006 cellular model implemented in the Oxford Chaste software. The extracellular cardiac surface potentials at each time frame were expanded into a Fourier series over the first N right singular vectors of the heart–torso transfer matrix of the inverse electrocardiography problem. Electrophysiological data were then computed for different values of N. Specifically, the localization of PVC foci (PVCF) and the latest electrical activation sites (LEAS) was performed using isopotential maps; fibrosis zones were identified using unipolar voltage maps; activation fronts were detected using the −du/dt max activation timing method; and cores of electrical rotors (CER) were identified by phase mapping based on the Hilbert transform.
Results. For PVCF, LEAS, and CER, N = 120 singular vectors were sufficient to achieve a localization error below 10 mm in 90% of cases. In contrast, accurate detection of activation wavefronts and fibrosis zones required more than 500 singular vectors (N > 500).
Conclusion. The "effective" dimension N of the solution space of the inverse electrocardiography problem (i.e., the required number of singular vectors) is a limiting factor for noninvasive cardiac imaging. The proposed spectral method is applicable for feasibility analysis of noninvasive mapping of the target electrophysiological process.