Towards Enabling Cardiac Digital Twins of Myocardial Infarction Using Deep Computational Models for Inverse Inference

TL;DR

This work tackles inverse inference of MI tissue properties from ECG within a multi-modal cardiac digital twin that fuses cine MRI and ECG data. It introduces anatomical and functional twinning to create subject-specific meshes and electrophysiological simulations, paired with a dual-branch variational autoencoder to predict MI location and distribution from simulated QRS. A comprehensive sensitivity analysis links infarct location, size, and transmurality to QRS changes and demonstrates a promising inverse inference performance, with Dice scores for scar and border zone segmentation and strong localization in transmural extensive anterior MI. The results support the feasibility of non-invasive, personalized CDT frameworks for MI, while acknowledging limitations and outlining paths toward clinical validation and more realistic torso and time-resolved modeling.

Abstract

Cardiac digital twins (CDTs) have the potential to offer individualized evaluation of cardiac function in a non-invasive manner, making them a promising approach for personalized diagnosis and treatment planning of my-ocardial infarction (MI). The inference of accurate myocardial tissue properties is crucial in creating a reliable CDT of MI. In this work, we investigate the feasibility of inferring myocardial tissue properties from the electrocardiogram (ECG) within a CDT platform. The platform integrates multi-modal data, such as cardiac MRI and ECG, to enhance the accuracy and reliability of the inferred tissue properties. We perform a sensitivity analysis based on computer simulations, systematically exploring the effects of infarct location, size, degree of transmurality, and electrical ac-tivity alteration on the simulated QRS complex of ECG, to establish the limits of the approach. We subsequently present a novel deep computational model, comprising a dual-branch variational autoencoder and an inference model, to infer infarct location and distribution from the simulated QRS. The proposed model achieves mean Dice scores of 0.457 \pm 0.317 and 0.302 \pm 0.273 for the inference of left ventricle scars and border zone, respectively. The sensitivity analysis enhances our understanding of the complex relationship between infarct characteristics and electrophysiological features. The in silico experimental results show that the model can effectively capture the relationship for the inverse inference, with promising potential for clinical application in the future. The code will be released publicly once the manuscript is accepted for publication.

Figures (11)

• Figure 1: The cardiac digital twin (CDT) generation workflow combining cine cardiac magnetic resonance images (MRIs) and electrocardiogram (ECG). Here, the anatomical twinning personalizes the geometrical model, while functional twinning personalizes the electrophysiological model. The anatomical and electrophysiological parameters include electrode positions, myocardial infarction (MI) distribution, ventricular muscle fiber orientation, Purkinje system, etc. Our goal is to solve the inverse problem for inferring the infarct location map (highlighted via the glow effect) from simulated QRS.
• Figure 2: The seven infarct locations defined on the 17-segment American Heart Association (AHA) model. The selection of the seven locations is referring to journal/CCR/nikus2014. Ext: extensive; Lim: limited.
• Figure 3: Illustration of several post-MI scenarios, including different infarct locations, sizes, and transmural extents. Here, scars refer to the area of damaged or dead heart muscle tissue that has been replaced by non-functional fibrous tissue, while the border zone (BZ) is the area surrounding the scar tissue where there may be some remaining damaged heart muscle tissue that is not yet fully scarred.
• Figure 4: Illustration of regional alterations in ventricular activation when scars are present in the heart. Here, we employ the subject with transmural extensive anterior MI as an example, to compare its activation time map (ATM) and QRS with that of a corresponding healthy one. The arrows highlight the areas where ATM differs in MI and healthy cases.
• Figure 5: Deep computational model for the inverse inference of MI based on a dual variational autoencoder (VAE). Note that the reconstructed point clouds (PCs) include both dense and sparse PCs and the simulated QRS includes 8 leads. For simplicity, the schematic of sparse PC is omitted, and only single lead is presented here.