Computational modelling for congenital heart disease: how far are we from clinical translation?

Abstract

Computational models of congenital heart disease (CHD) have become increasingly sophisticated over the last 20 years. They can provide an insight into complex flow phenomena, allow for testing devices into patient-specific anatomies (pre-CHD or post-CHD repair) and generate predictive data. This has been applied to different CHD scenarios, including patients with single ventricle, tetralogy of Fallot, aortic coarctation and transposition of the great arteries. Patient-specific simulations have been shown to be informative for preprocedural planning in complex cases, allowing for virtual stent deployment. Novel techniques such as statistical shape modelling can further aid in the morphological assessment of CHD, risk stratification of patients and possible identification of new 'shape biomarkers'. Cardiovascular statistical shape models can provide valuable insights into phenomena such as ventricular growth in tetralogy of Fallot, or morphological aortic arch differences in repaired coarctation. In a constant move towards more realistic simulations, models can also account for multiscale phenomena (eg, thrombus formation) and importantly include measures of uncertainty (ie, CIs around simulation results). While their potential to aid understanding of CHD, surgical/procedural decision-making and personalisation of treatments is undeniable, important elements are still lacking prior to clinical translation of computational models in the field of CHD, that is, large validation studies, cost-effectiveness evaluation and establishing possible improvements in patient outcomes.

Figure 1

An example of patient-specific simulation for virtual device implantation in the right ventricular outflow tract. The patient-specific anatomy is reconstructed from cardiovascular magnetic resonance (CMR) imaging, taking into account deformations over the cardiac cycle (ie, whole heart configuration in diastole vs systolic configuration). All available devices can be implanted virtually, including simulation of prestenting. Computational analyses can then offer predictions, for example, stresses exerted by the different devices on the vessel wall, aiding in the decision-making process.

Figure 2

An example of patient-specific simulation for coarctation stenting, showing virtual device positioning (A) and deployment (B), as well as the corresponding fluoroscopy data (C and D) acquired in vivo.

Figure 3

Summarising a statistical shape analysis framework (SSM, statistical shape model). Starting from segmentation of medical imaging data (1), the anatomical segment of interest, for example, the left ventricle, is reconstructed (2) and these two steps are repeated for all the patients in the population, generating the inputs for computing the mean shape (3); postprocessing (4) involves methods such as principle component analysis (PCA), allowing to compare variations in shape (eg, ±2 SD from the mean shape) and perform quantitative assessments.