Page 20 - Artificial Intelligence for Computational Modeling of the Heart

xx Foreword

Figure 0.1. How AI and computational modeling can push the knowledge frontier
for precision medicine.

visible data with physiological knowledge. For instance, fractional
flow reserve (FFR) can now be estimated from CT to identify phys-
iologically critical stenoses in coronary arteries. MR-based meth-
ods for target identification in complex arrhythmia such as atrial
fibrillation and ventricular tachycardia are becoming a reality.
Patient-specific modeling has therefore the potential to augment
imaging with quantitative features of heart physiology and its un-
derlying biophysics, mathematically represented as time-varying
vector or tensor fields.
With the increasing availability of very large medical database
comprising imaging, reports, patient record, etc., artificial intelli-
gence (AI) has the potential to disrupt healthcare as we know it,
similar to what we are witnessing in other industries. AI consists
in learning complex models directly from large databases. Deep
learning, in particular, unlocks the discovery of data features and
patterns that allow higher levels of predictive performance com-
pared to traditional machine learning approaches. AI they already
allowed unprecedented success in various applications, such as
cardiac motion analysis for survival prediction and human-level
arrhythmia detection in electrocardiogram, to cite just a few ex-
amples. Nevertheless, AI algorithms, especially deep learning, are
hard to examine after the fact to understand specifically how and
why a decision has been made. This lack of explainability of AI
hampers our ability to improve mechanistic understanding of the
disease and develop effective therapies. For example, AI can pick
up the features of low cardiac function in 12-lead electrocardio-
gram that were previously unrecognized, but it remains unknown