Page 230 - Artificial Intelligence for Computational Modeling of the Heart

Chapter 6 Additional clinical applications 203

output of a computational blood flow model, it inherits the limi-
tations of the blood flow model. The definition of the parameters
of the blood flow models (both reduced-order and 3D CFD) relies
on physiological assumptions, which would also require valida-
tion on larger data sets. Such assumptions include for example the
mechanical behavior of the vascular wall (rigid, elastic, viscoelas-
tic), the use of population-averaged rheological properties of the
blood, and so on.

6.3 Whole-body circulation

6.3.1 Introduction
Due to the prohibitive computational cost of spatially de-
tailed blood flow models (three-dimensional models in particu-
lar), closed loop models of the cardiovascular system rely heavily
on lumped parameter modeling techniques, which are based on
the analogy between hydraulic and electrical systems. The whole
body circulation (WBC) model employed herein, displayed in
Fig. 6.11, contains a heart model (left ventricle (LV) and atrium,
right ventricle and atrium, valves), the systemic circulation (ar-
teries, capillaries, veins), and the pulmonary circulation (arter-
ies, capillaries, veins) [453]. Chapter 3 has described established
techniques for cardiac image parsing and structure tracking. The
anatomical information extracted using these methods, e.g. time-
varying ventricular volume, may then be used as input for blood
flow computations, as described more extensively in section 2.4.
While the lumped parameter model is computationally very ef-
ficient (a single forward run requires a runtime of milliseconds),
its personalization requires hundreds of forward runs, leading to
an overall computation time of 30 to 60 seconds for determin-
ing the patient-specific measures of interest. To remove the need
for that computationally intensive step, we propose to estimate
those measures using a deep neural network. This approach leads
to a significant acceleration of computation time, thus making
this modeling framework suitable for use in clinical workflows
even within an intra-operative setting, where decisions need to be
taken in near real-time.

6.3.2 Methods
6.3.2.1 Traditional personalization framework
The WBC model may be run under patient-specific conditions
to compute various clinically relevant measures of interest: arte-
rial resistance, arterial compliance, dead volume of the left/right