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

Area of Science:

• Computational Biology and Medicine
• Cardiovascular Research
• Machine Learning in Healthcare

Background:

• Myocardial infarction (MI) remodeling increases arrhythmia risk, necessitating personalized risk prediction.
• Patient-specific computational models show promise but are time-intensive for clinical use.
• Machine learning (ML) and neural networks (NN) can enhance prediction accuracy for arrhythmias.

Purpose of the Study:

• To develop and evaluate an integrated image-based, patient-specific computational simulation and ML methodology for assessing arrhythmia risk in post-MI patients.
• To demonstrate the efficacy of simulation-supported data augmentation in improving the accuracy of arrhythmia prediction models.

Main Methods:

• Constructed MRI-based computational models from 30 post-MI patients (baseline population).
• Augmented the patient population by creating geometric variations to expand the dataset (augmented population).
• Used ML (k-NN, SVM, logistic regression, xgboost, decision tree) and NN (MLP) models trained on geometric features (myocardial/ischemic volume, segment percentages) to predict simulation outcomes (arrhythmia/no-arrhythmia).

Main Results:

• The augmented population (129 models) showed a reduced incidence of reentry (13.0%) compared to the baseline (21.8%).
• ML and NN models achieved higher prediction accuracy on the augmented population (0.88-0.89) compared to the baseline (0.83-0.86).
• Simulation-supported data augmentation significantly improved the predictive performance of ML and NN models.

Conclusions:

• Combined patient-specific simulations and ML offer a rapid, accurate, and efficient method for clinical insights into arrhythmia risk.
• Simulation-supported data augmentation is a valuable strategy to enhance predictive models, especially with limited patient data.
• This data-driven simulation approach holds potential for predicting dangerous arrhythmias in post-MI patients.