Page 218 - Artificial Intelligence for Computational Modeling of the Heart

Chapter 6 Additional clinical applications 191

consists in performing blood pressure measurements in the upper
and lower body extremities, and computing the difference [428].
As described in section 2.4, a different approach relies on Com-
putational Fluid Dynamics (CFD) for performing patient-specific
blood flow computations. Compared to traditional techniques,
the main advantage of a CFD based tool is that it offers the poten-
tial to also simulate post-treatment hemodynamics or flow under
different conditions such as stress, which would normally have to
be induced pharmacologically. Furthermore CFD is able to pro-
vide quantities that are not available through measurement, e.g.
Wall Shear Stress (WSS) or WSS-derived quantities. As comput-
ing power and technology improved over the years, CFD has been
used increasingly as a research tool for studying aortic flow under
patient-specific conditions and different physiological assump-
tions: rigid vessel walls [429,430] or compliant walls [431–433],
personalized inlet boundary conditions [434], idealized [433]or
reduced order CFD modeling [435,436], etc.
CFD based approaches were also proposed for assessing aortic
coarctation patients. These methods typically rely on medical im-
ages acquired through either magnetic resonance imaging (MRI)
[437] or Computed Tomography (CT) [229], in addition to some
physiological parameters, such as cuff blood pressures or echocar-
diographic flow velocities.
Ituetal. [436] proposed a non invasive workflow, based on
MRI data, for hemodynamic assessment of aortic coarctation, re-
lying on a reduced-order CFD model coupled with a pressure drop
model for the CoA segment. The pressure drop model was based
on the work previously published by Young and Tsai [438], and in-
tegrated into the reduced-order model for accurate computation
of energy losses at the CoA site. However, the pressure drop model
was designed and developed based on idealized stenotic shapes,
which do not reflect the complexities and anatomical variabilities
of aortic coarctation.
As machine learning (ML) evolved and was successfully ap-
plied to solving more complex problems, it also gained popular-
ity as a technique for accelerating/replacing complex computa-
tions that would normally require long execution times such as
computational fluid or structural dynamics. Liang et al. [439]pro-
posed a machine learning approach for replacing computation-
ally demanding Finite Element Method (FEM) computations for
predicting surface stress distribution in aortic vessels. Specifically,
they performed a large number of FEM computations on syn-
thetically generated aortic vessels, and then trained a neural net-
work to directly predict the surface stress distribution using only
the vessel shape as input. Similarly, Liang et al. [440] proposed