Page 212 - Artificial Intelligence for Computational Modeling of the Heart

Chapter 6 Additional clinical applications 185

were analyzed by radiologists to determine the scar burden per
region of a 17 segment left-ventricular model as recommended by
the American Heart Association (AHA). The affected AHA regions
were hereby categorized based on scar characteristics as either in-
tramural, epicardial, diffuse, or transmural scar and assessed at
25%, 50%, 75%, or 100%.
During the intervention, the AHA segment associated with the
location of the left ventricular electrode and the septal location
(apical, mid, or basal) of the RV electrode was reported [206].
Atrio-ventricular (AV) pacing delay and VV delay information was
available additionally.
Three different configurations, namely LV only, RV only, or
biventricular (BiV) stimulation, were tested in all patients, except
one, resulting in a total of 28 stimulation protocols. The AV and VV
delay per stimulation protocol was kept constant per patient. For
validation, 12-lead ECGs were acquired.

6.1.2.2 Computational modeling
A personalized multi-physics computational model was esti-
mated using the pipeline illustrated in Fig. 6.1.First theheart
anatomy was segmented from the MR images as described in sec-
tion 3.2 and refined in cases of under- and over-segmentation [31].
The endo- and epicardial segmentations given as surface triangu-
lations were merged to form a closed surface and then tetrahedral-
ized, resulting in a volumetric bi-ventricular myocardial model.
A rule-based fiber model and mesh tagging was applied (refer to
section 2.1). Furthermore, the scar burden per AHA segment, as
per radiological report, was mapped to the model. Finally, a repre-
sentative avatar of the human torso was fitted to the images, and
ECG electrodes were placed consistently across patients accord-
ing to clinical standards (section 2.1).
To compute the electrophysiological (EP) response, the graph-
based model described in section 2.2.3 was applied. Left bundle
branch block (LBBB) was modeled by disabling the associated
left activation patway and constraining the left Purkinje system to
have a conduction velocity equal or at most within 125% of the
conduction velocity in the myocardium. The EP model was indi-
vidualized following the approach presented in section 2.5.2.
For each patient the outcome from CRT treatment was pre-
dicted by simulating the stimulation protocols that had been pre-
viously tested on the patient. Left and right electrodes were placed
as stated in the clinical report. Heart rate and pacing delays were
also set accordingly. To cope with the uncertainty in the timing of
the stimulations with respect to the spontaneous activation of the
heart, which could potentially affect the presence of wave fusion,