Key result
Hybrid biophysical-neural network model simulates cardiac electrophysiology ~10,000-fold faster than standard finite element methods.
Why the study?
Computationally expensive models resolving intercalated disk nanostructure have been confined to cables of tens of cells, leaving the role of intercalated disk nanostructure on cardiac tissue conduction untestable.
Does a hybrid biophysical-neural network surrogate model accurately reproduce cardiac action potentials and conduction velocity restitution while reducing computational cost compared to finite element method simulations in 1D, 2D, and 3D simulated cardiac tissues?
Does a hybrid biophysical-neural network surrogate model accurately reproduce cardiac action potentials and conduction velocity restitution while reducing computational cost compared to finite element method simulations in 1D, 2D, and 3D simulated cardiac tissues?
This hybrid biophysical-neural network model enables tissue-scale 2D and 3D cardiac conduction simulations that resolve nanoscale intercalated disk remodeling at speeds 10^4 to 10^5 times faster than finite element models, demonstrating that combined reduced gap-junctional coupling and regional cleft widening cause arrhythmogenic conduction block.
Answers are grounded in this paper and cite their sources.
Enables scalable multiscale modeling of intercalated disk effects; leaves open clinical relevance for conduction abnormalities.
Nanoscale structure of the intercalated disk (ID) governs cardiac conduction through gap junctions and ephaptic interactions and is remodeled in arrhythmogenic disease. Computationally expensive models that resolve ID nanostructure have been confined to cables of tens of cells, such that the role of ID nanostructure on conduction has not been testable. Here, we show that a hybrid biophysical-neural network model removes this limit. A memory-based flow map, trained on finite element method (FEM) simulations of single junctions and short cables, defines surrogates that replace each junction in tissue. The hybrid model reproduces action potentials with 1-2 mV RMSE and conduction velocity restitution across multiple ID geometries with a mean error of 0.24%. Without retraining, surrogate models run in 2-D and 3-D tissues, with over one million junctions, 10^4-10^5 times faster than the FEM model and within an order of magnitude of a monodomain model that neglects ID structure. In large tissues with reduced gap-junctional coupling, regional cleft widening blocks conduction, whereas either change alone does not, linking ID remodeling to arrhythmogenic substrate.
No takes yet. Share an insight, caveat, or question.
Sui et al. (2026) studied Cardiac conduction and arrhythmias. Hybrid biophysical-neural network surrogate model (memory-based flow map) vs. Finite element method (FEM) reference model and standard monodomain model was evaluated on Action potential root mean square error (RMSE), conduction velocity restitution error, and computational speed-up. The hybrid biophysical-neural network model reproduced cardiac action potentials with 1-2 mV RMSE and conduction velocity restitution with 0.24% mean error while running 10^4-10^5 times faster than FEM.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: