Page 22 - Artificial Intelligence for Computational Modeling of the Heart

Preface xxiii

Preface

What if we could create a digital representation of a person’s heart,
that functions and beats like the real organ? What if the virtual
heart could react to physiological stresses, diseases, and interven-
tions as if real? This concept, known as digital twin, is well estab-
lished in the world of manufacturing and industry [1]. Many engi-
neering artifacts, from airplane to cars to highly complex wind tur-
bines, among others, are first developed numerically. The result-
ing digital twins are nowadays commonly used to predict failure,
plan maintenance strategies, or even guide operations in a chang-
ing environment [2]. Digital twins are also used to train robotic
systems powered by artificial intelligence (AI), before deploying
them in the field [3]. With the increasing cost of physical testing,
due to the increasing complexity of the manufactured systems in
particular, it is not a surprise that digital twins are becoming a key
engineering tool in manufactory and industry.
What if that same concept would be possible in healthcare?
One could use a digital twin of a person to predict disease, derive
personalized risk scores and allow patients to manage their health
proactively, enable physicians to optimize therapies for specific
patients, and much more. Digital twins could become a key com-
ponent of the new wave of precision medicine [4–6].
Estimating the digital twin of a physical artifact is not a sim-
ple task [1]. First, it requires designing and implementing accurate
models of the physics underlying the function of the artifact. In
industry, the artifact is created from scratch and directly designed
using computer assisted design software. Its physics, for instance
mechatronics, is therefore known by design. Second, the compu-
tational model of the artifact needs to be individualized. In other
words, the parameters of the model need to be adjusted such that
the digital twin replicates how the artifact of interest works along
with its state, including potential defects in the object, the effects
of previous events, and physical properties. The individualization
task needs to be performed continuously, as data from sensors
flow, to keep the digital twin up to date. Finally, interactions with
the environment need to be accounted for, such as the effects of
natural elements, human interactions, or simply long term wear
of the artifact materials. This step is crucial if one wants to use the
digital twin for predictive maintenance, error recovery or simply
operational optimization [2].
Creating digital twins of biological systems can be daunting.
Contrary to industrial artifacts, the underlying biophysics are still