Digital twinning of cardiac electrophysiology for congenital heart disease

Abstract

In recent years, blending mechanistic knowledge with machine learning has had a major impact in digital healthcare. In this work, we introduce a computational pipeline to build certified digital replicas of cardiac electrophysiology in paediatric patients with congenital heart disease. We construct the patient-specific geometry by means of semi-automatic segmentation and meshing tools. We generate a dataset of electrophysiology simulations covering cell-to-organ level model parameters and using rigorous mathematical models based on differential equations. We previously proposed Branched Latent Neural Maps (BLNMs) as an accurate and efficient means to recapitulate complex physical processes in a neural network. Here, we employ BLNMs to encode the parametrized temporal dynamics of in silico 12-lead electrocardiograms (ECGs). BLNMs act as a geometry-specific surrogate model of cardiac function for fast and robust parameter estimation to match clinical ECGs in paediatric patients. Identifiability and trustworthiness of calibrated model parameters are assessed by sensitivity analysis and uncertainty quantification.

Keywords: neural maps; numerical simulations; parameter estimation; single ventricle physiology; uncertainty quantification.

Figure 1.

Sketch of the computational pipeline. We reconstruct the patient-specific geometry with HLHS from imaging. We generate a dataset of electrophysiology simulations encompassing cell-to-organ variability in model parameters. We train a BLNM that effectively reproduces 12-lead ECGs while covering model variability. We employ the BLNM for digital twinning.

Figure 2.

Comparison of the segmentations and the reconstructed anatomic models and meshes between the HLHS patient and a healthy patient with normal cardiac anatomy.

Figure 3.

Physics-based electrophysiological modelling dataset generation. Full dataset containing 200 in silico precordial and limb leads recordings (blue, solid) and patient-specific 12-lead ECGs (black, dashed). ECG, electrocardiogram.

Figure 4.

Physics-based electrophysiological modelling. Spatio-temporal membrane action potential evolution for one electrophysiology simulation in the dataset performed on the HLHS paediatric patient. HLHS, hypoplastic left heart syndrome.

Figure 5.

Sensitivity analysis for the seven model parameters encoded in the BLNM via Shapley effects.

Figure 6.

Inverse uncertainty quantification: parameter uncertainty. One-dimensional views of the posterior distribution. Different colours represent different HMC chains.

Figure 7.

Inverse uncertainty quantification: matching clinical data. Clinical recordings (dashed, black) and mean estimation (red, solid) of 12-lead ECGs for the HLHS paediatric patient via HMC. Light red encompasses the variability between mean ± 5 s.d.

Figure 8.

Running in silico disease scenarios. Activation (top) and repolarization (bottom) maps with personalized model parameters (left), left bundle branch block (centre) and right bundle branch block (right).