Page 71 - Artificial Intelligence for Computational Modeling of the Heart

Chapter 1 Multi-scale models of the heart for patient-specific simulations 41

carefully generating synthetic training databases using computa-
tional models can enable effective training schemes even when
real data is scarce. Genetic algorithms have also been investigated
as ways to estimate model parameters from clinical data [129,
200]. In particular, such algorithms were combined with a multi-
fidelity modeling approach, where a detailed 3D model was ap-
proximated by an efficient 0D surrogate model to speed up the
overall estimation process of active and passive tissue parame-
ters [129].

1.5.4 Stochastic methods
Finally, stochastic methods aim to find the “optimal” parame-
ter set along with their confidence values, considering uncertain-
ties in data and models [26,82,201].

1.5.5 Streamlined whole-heart personalization
Comprehensive and fast 3D whole-heart model personaliza-
tion is nowadays becoming possible. In [18], the authors pre-
sented a first approach. However, it involved significant manual
steps and its robustness to different patients, pathologies and
data quality remains an open question. To our knowledge, only
few comprehensive, integrated pipelines have been presented so
far [6,9] to personalize anatomy, electrophysiology, biomechanics
and hemodynamics in a streamlined, consistent and automatic
fashion. Section 2.5 will present in details one such framework.

1.6 Summary
This chapter introduced models and approaches for multi-
physics, multi-scale modeling of heart function. Several modeling
options are available, going from very detailed models to coarse
ones. Like in other scientific domains, it is likely that one model
would not fit all the clinical questions that could be explored and
potentially answered using multi-physics approaches. It is there-
fore the task of the researchers, modelers, computer scientists and
physicians together, to identify 1) the right level of model details
to use, 2) what computational framework and numerical approxi-
mations to afford and 3) which parameters should be estimated
for personalized simulations. One strategy that proved useful is
to start from the clinical challenge to address, and answer these
questions accordingly. This was the approach used for all exam-
ples described in this book, resulting in a wide variety of models,
from simple to complex, but all adjusted to the clinical workflow