Impact of Geometric Simplifications on Cardiac Computational Models of Electrophysiology to Assess Cardiac Resynchronization Therapy

Cardiac computational models are generally based on in vivo patient imaging data, often magnetic resonance imaging (MRI), with a typical resolution of 0.7 × 0.7 × 10 mm. As a consequence, they lack endocardial structures like trabeculae and false tendons. Furthermore, the reconstruction of the basal and apical regions from images is impossible due to very high slice thickness and spacing; thus, it is either omitted or managed using prior information about the heart shape. Cardiac computational models have been frequently employed to assess responses to cardiac resynchronization therapy (CRT). CRT is a widely used therapy in heart failure patients with moderate to severe dyssynchrony; however, a large number of patients are reportedly classified as nonresponders (around 30%). Left bundle branch block (LBBB) is one of the primary conditions that require CRT. Numerous studies have been devoted to cardiac computational models to better identify patient candidates for this therapy and to optimize lead placement(s). The objectives of our work are to create computational models of electrophysiology of human cardiac anatomies with a range of detail, reproduce a LBBB population of patients with and without CRT devices, and analyze the impact of losing anatomical detail due to image resolution on CRT simulations by comparing them to highly detailed anatomies. Our analysis was performed on four anatomically normal human hearts extracted from ex vivo high-resolution MR images. Each detailed heart was adapted to replicate commonly used models: (1) the same model with trabeculations, but cropped just under the valves; (2) a model with smooth walls; and (3) a model with smooth walls, cropped just under the valves. For each of the above models (4 hearts × 4 detail levels), we obtained results for four different heart conditions: healthy, LBBB, LBBB with CRT epicardial lead placement, and LBBB with CRT endocardial lead placement. The ventricular cardiomyocyte model of O’Hara-Rudy was solved using a monodomain approximation of electrophysiological propagation. Normal distribution of fiber orientations was employed in all anatomies using the published model of Doste, and activation locations were set following the work of Durrer. Results show a large variability of pseudo-electrocardiogram (ECG) morphology, QRS duration, and last activated region between all models. Most of the differences between the simulations arise from additional pathways of electrical conduction created by trabeculations, causing differences in local depolarization patterns, with either shortened or prolonged QRS complexes: smoothed-incomplete geometries demonstrated a 9 ms shorter QRS than the detailed anatomies and a mean distance of 1.23 cm between the last activated point for LBBB hearts. Geometric simplifications led to different activation patterns, incorrect QRS characterization, and differences in pseudo-ECG morphology, thus hampering accurate CRT planning; these simplifications should be taken into consideration when the aim of the model is to reproduce clinically measured ECG morphology. The small population considered in this work did not allow us to establish clear patterns, and more hearts need to be analyzed. Additionally, the geometric simplifications provided disperse CRT recovery responses from the model, a finding that should also be taken into account when optimizing CRT therapy from in silico studies.