Hierarchical multiple-model Bayesian approach to transmural electrophysiological imaging

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Noninvasive electrophysiological (EP) imaging reconstructs cardiac current sources from electrocardiographic (ECG) data.
• This process is ill-posed and typically uses fixed models, which may not match real source distributions.
• Existing methods struggle with varying spatiotemporal source dynamics.

Purpose of the Study:

• To develop a novel hierarchical Bayesian approach for transmural EP imaging.
• To overcome the limitations of fixed constraining models in cardiac EP imaging.
• To accurately reconstruct cardiac current sources with diverse spatial and temporal characteristics.

Main Methods:

• Proposed a hierarchical Bayesian framework for EP imaging.
• Employed a continuous combination of multiple spatial models using an Lp-norm prior.
• Treated the Lp-norm parameter 'p' as an unknown hyperparameter with a prior distribution.
• Inferred the optimal weighting of models from ECG data via posterior distribution of 'p'.

Main Results:

• The proposed method demonstrated consistent performance in reconstructing cardiac sources.
• Successfully reconstructed sources with various extents and structures, unlike fixed L1- and L2-norm models.
• Validated accuracy through synthetic and real-data experiments on human heart-torso models.

Conclusions:

• The hierarchical Bayesian approach offers a flexible and accurate solution for noninvasive cardiac EP imaging.
• This method adapts to varying source distributions, improving upon fixed model limitations.
• Provides a robust tool for understanding cardiac electrical activity noninvasively.