Robust Transmural Electrophysiological Imaging: Integrating Sparse and Dynamic Physiological Models into ECG-Based Inference

Jingjia Xu, John L. Sapp, Azar Rahimi Dehaghani, Fei Gao, Milan Horacek, Linwei WangMICCAI 2015 · OpenReview

Abstract

Noninvasive inference of patient-specific intramural electrical activity from surface electrocardiograms (ECG) lacks a unique solution in the absence of prior assumptions. While 3D cardiac electrophysiological models emerged to be a viable vehicle for constraining this inference with knowledge about the spatiotemporal dynamics of cardiac excitation, it is important for the inference to be robust to errors in these high-dimensional model predictions given the limited ECG data. We present an innovative solution to this problem by exploiting the low-dimensional structure of the solution space – a powerful regularizer in overcoming the lack of measurements – within the dynamic inference guided by physiological models. We present the first Bayesian inference framework that allows the exploration of both the spatial sparsity of cardiac excitation and its complex nonlinear spatiotemporal dynamics for an improved inference of patient-specific intramural electrical activity. The benefit of this integration is verified in both synthetic and real-data experiments, where we present one of the first detailed, point-by-point comparison of the reconstructed electrical activity to in-vivo catheter mapping data.

Introduction

Reconstructing the 3D electrical activity inside the heart wall (transmural) from electrodes on the body surface (ECG) is a notoriously ill-posed inverse problem. Because surface data is limited and many different internal electrical patterns can produce the same surface ECG, researchers must rely on prior assumptions to find a unique solution. While 3D cardiac electrophysiological (EP) models provide rich domain knowledge about how excitation spreads, these models are high-dimensional and often inaccurate due to patient-specific variations or simplified parameters. If the inference relies too heavily on a flawed model, the results will be wrong. Conversely, purely data-driven sparse methods often fail to capture the complex temporal dynamics of a heartbeat. This paper seeks to bridge that gap by integrating these two complementary priors.

Methods

The proposed hierarchical Bayesian framework decomposes the intramural action potential ($u_k$) into two independent components: a spatial profile ($s_k$) and a temporal profile ($t_k$).

Hierarchical Bayesian framework and inference flow. (a) shows the model structure where the action potential is influenced by both dynamic predictions and a sparsity model. (b) details the recursive Variational Bayesian (VB) inference process.

The Physiological Dynamic Prior ($t_k$) uses the Aliev-Panfilov model to simulate electrical wave propagation. To handle high dimensionality and nonlinearity, a simulation-based approach using the Unscented Transform is employed. The Sparse Prior ($s_k$) applies Total Variation (TV) regularization, exploiting the fact that heart electrical states have sharp gradients between active and inactive regions. Because the resulting posterior is intractable, Variational Bayesian Inference is used to find an approximate solution, allowing the system to recursively update its estimate as new ECG data arrives.

Results and Validation

The method was tested on synthetic human heart-torso models and a real-patient pilot study. Synthetic tests included myocardial infarction cases where the model was intentionally "wrong" (using standard parameters for a scarred heart).

Comparison of reconstructed electrical activity. The integrated method (Integration) significantly outperforms purely dynamic or sparse models, maintaining high correlation with ground truth even when the wavefront encounters unmodeled infarct scars.

As illustrated in Figure 4, the integrated method maintained high accuracy even when the wavefront encountered unmodeled infarct scars. In the real-patient study, the noninvasive reconstruction correctly identified low-voltage "scar" regions. Statistical analysis showed a significant negative association ($p = 0.0002$) between reconstructed activation time and bipolar voltage, matching clinical expectations for infarcted tissue.

Discussion and Conclusion

The integration of a sparse prior acts as a safety net, forcing the solution to respect the structure of actual ECG data when the dynamic physiological model makes errors. This "plug-and-play" framework is flexible, allowing any cardiac model to be used. While the pilot study is promising, future work will involve larger clinical trials and accounting for the mechanical motion of the heart during the cardiac cycle.

Summary updated

Abstract

Noninvasive inference of patient-specific intramural electrical activity from surface electrocardiograms (ECG) lacks a unique solution in the absence of prior assumptions. While 3D cardiac electrophysiological models emerged to be a viable vehicle for constraining this inference with knowledge about the spatiotemporal dynamics of cardiac excitation, it is important for the inference to be robust to errors in these high-dimensional model predictions given the limited ECG data. We present an innovative solution to this problem by exploiting the low-dimensional structure of the solution space – a powerful regularizer in overcoming the lack of measurements – within the dynamic inference guided by physiological models. We present the first Bayesian inference framework that allows the exploration of both the spatial sparsity of cardiac excitation and its complex nonlinear spatiotemporal dynamics for an improved inference of patient-specific intramural electrical activity. The benefit of this integration is verified in both synthetic and real-data experiments, where we present one of the first detailed, point-by-point comparison of the reconstructed electrical activity to in-vivo catheter mapping data.
BibTeX
@inproceedings{xu2015robust,
  author = {Jingjia Xu and John L. Sapp and Azar Rahimi Dehaghani and Fei Gao and Milan Horacek and Linwei Wang},
  title = {{Robust Transmural Electrophysiological Imaging: Integrating Sparse and Dynamic Physiological Models into ECG-Based Inference}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2015},
  url = {https://openreview.net/forum?id=MJprrEhnTlr},
}
Figure