Page 23 - Artificial Intelligence for Computational Modeling of the Heart

xxiv Preface

not fully understood. The high fidelity physiological models that
have been so far are highly non-linear, with multiple feedback
loops and controls, spanning spatio-temporal scales going from
the molecules to the body, from nano seconds to many years [7].
Furthermore, biological data can require invasive or destructive
measurement procedures, they may be noisy, or too sparse and in-
complete. Despite these challenges, tremendous progress towards
patient-specific modeling of organ and system function has been
made in the past decades, with initial clinical proof of concepts
and validation already happening [5,6,8,9]. The heart is one of
the organs that has attracted significant interest from biophysiolo-
gists, mathematicians and modelers. Researchers first used math-
ematical models to understand heart function and the underly-
ing physiology, progressively moving towards clinical applications
like the characterization of heart failure. Evolution of technology,
computational power and healthcare sensing technologies (e.g.
imaging scanner, electrophysiology sensors, etc.) has been pivotal
towards this progression [4].
In parallel, we are witnessing a new wave of artificial intel-
ligence developments [10]. Fueled by the exponential increase
of available digital data and computational power, new methods
based on deep neural networks are being developed, disrupting
a wide range of fields, from computer vision to natural language
processing, and now including computational biology, physics
and healthcare applications. Provided a large amount of repre-
sentative data is available, deep learning algorithms learn with
unprecedented effectiveness the feature patterns that are most
relevant for a specific task (e.g. classification or prediction), reach-
ing unparalleled levels of performance [11].
Motivated by these two trends, the rise of digital twins and ar-
tificial intelligence, this book presents opportunities arising from
the combination of both fields. We believe that the performance
achieved by AI systems, informed by biophysical knowledge and
models, could provide a new generation of tools for improved
health management, with potential impact on increased favorable
treatment outcomes and clinical workflow optimization. While
traditional computational models of the heart are complex to
use, requiring specific expertise to build and manipulate them,
AI could foster a new way of integrating available data and phys-
iological models. Specific modeling components could be built
automatically, such as the anatomical model that is created from
multi-modality imaging data. Modeling tasks could be made dra-
matically more computationally efficient, such as the estimation
of individualized model parameters or the prediction of quan-
tities of clinical interest. Additional and more powerful feature