Page 29 - Artificial Intelligence for Computational Modeling of the Heart

xxx Preface

view of cardiac electrophysiology. The inverse problem of param-
eter estimation is also often ill-posed. Traditional approaches rely
on optimization techniques with ad hoc regularizers or optimiza-
tion strategies. This chapter explores how machine learning meth-
ods can be used to solve the inverse problem. A first approach
consists in learning a regression model that maps clinical mea-
surements to model parameters directly using a large database of
synthetic data. The method is illustrated for the case of cardiac
electrophysiology. A second approach builds upon reinforcement
learning techniques to learn the optimization process itself, lead-
ing to a potentially more efficient and more robust algorithm than
black box optimization methods.

Chapter 6: Additional clinical applications. The last chapter of
the book presents three additional clinical applications of the
methods presented in the previous chapters. First, a multi-scale,
multi-physics cardiac model for cardiac resynchronization ther-
apy (CRT) is presented. CRT is an effective treatment of heart
failure. It consists in placing an advanced pacemaker that stim-
ulates both right and left ventricles to resynchronize the failing
heart. Unfortunately, the therapy response rate is still low. Meth-
ods that could guide the electrophysiologist on where to implant
the stimulating electrodes and how to program them are needed.
In the presented work, the model is first built from pre-operative,
standard-of-care data. It is then used to simulate various pacing
protocols. Prediction results are compared with observed changes
in cardiac electrophysiology on ten patients illustrating the use of
the virtual heart to visualize the effects of the treatment before
delivering it. The second application relates to aortic coarctation
(CoA), a congenital disease that requires surgical treatment. The
decision to treat is based on the pressure drop across the coarc-
tation. This is assessed using an invasive procedure, with pressure
catheters. We present a method that allows the computation of the
pressure drop directly from images by learning a model from data
generated using computational fluid dynamics. Finally, the third
application addresses the entire cardiovascular system, based on
a lumped parameter model (LPM) of whole body circulation. A
similar approach to the one employed for the CoA use case is con-
sidered, relying on a purely synthetic training database. Both time
independent, e.g. systemic resistance and compliance, and time
dependent quantities, e.g. ventricular pressure, are predicted. We
show that the performance of the AI models is statistically similar
to that of the LPM, while the inference time is reduced to millisec-
onds.