AUTHOR=Maleckar Mary M. , Myklebust Lena , Uv Julie , Florvaag Per Magne , Strøm Vilde , Glinge Charlotte , Jabbari Reza , Vejlstrup Niels , Engstrøm Thomas , Ahtarovski Kiril , Jespersen Thomas , Tfelt-Hansen Jacob , Naumova Valeriya , Arevalo Hermenegild 

TITLE=Combined In-silico and Machine Learning Approaches Toward Predicting Arrhythmic Risk in Post-infarction Patients

JOURNAL=Frontiers in Physiology

VOLUME=Volume 12 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2021.745349

DOI=10.3389/fphys.2021.745349

ISSN=1664-042X

ABSTRACT=Post myocardial infarction (MI) remodeling significantly increases
patient arrhythmic risk. Simulations using patient-specific models have
shown promise in predicting personalised risk for arrhythmia. However,
these are computationally- and time-intensive, hindering
translation to clinical practice. Classical machine
learning as well as neural network techniques can be used to predict occurrence of arrhythmia as
predicted by simulations based solely on infarct and ventricular
geometry. We present an initial combined image-based patient-specific in silico and machine learning methodology to assess risk for dangerous arrhythmia in post-MI patients. Furthermore, we aim to demonstrate that simulation-supported data augmentation improves prediction models, combining patient data, simulation, and advanced statistical modeling to improve overall accuracy for arrhythmia risk assessment.
MRI-based computational models were constructed from 30 patients 5-days post-MI ("baseline"). 
To assess the utility biophysical model-supported data augmentation for improving arrhythmia prediction, we augmented the virtual baseline patient population. Each patient ventricular and scar geometry in the baseline population was used to create a subfamily of geometric models, resulting in an expanded patient population ("augmented"). 
Arrhythmia induction was attempted via programmed stimulation at 17 sites for each virtual patient
corresponding to AHA LV segments. Simulation outcome, "arrhythmia''/"no-arrhythmia'', was used as ground truth for machine learning models. For each patient geometric model, we measured choice data features:
myocardial and scar volumes, and segment-specific myocardial and scar percentages, as input to machine learning algorithms. We trained five classical and two neural network approaches to predict simulation outcomes from these geometric features alone.
Models were trained on 70% of randomly selected
segments and the remaining 30%  used for validation for both baseline and augmented populations. 
Stimulation in the baseline population (30 patient models) resulted in reentry in 21.4% of sites; in the augmented population (129 patient models) reentry occurred in 12.9%.
Models accuracy ranged from 83-86% for the baseline population, improving to 88-89% for all models (augmented).
Machine learning combined with patient-specific simulations can provide key clinical insights with accuracy,
rapidly and efficiently. Simulation-supported data augmentation can be employed to further improve predictive results. This work paves the way for data-driven simulations to
predict of dangerous arrhythmia in MI patients.