Abstract

Cardiac physiology results from coordinated interactions across multiple scales, from proteins through to whole heart function. Physics-based computational models can encode these multi-scale processes. Calibrating these models to clinical data measuring whole heart function potentially provides a virtual heart assay to identify the tissue and cellular scale mechanisms underpinning these clinical observations. However, these models have large numbers of parameters and are computationally intensive, making model calibration challenging. In this context, machine learning methods for parameter reduction and estimation can be of great help in reducing the number of required simulation runs and therefore make parameter fitting possible.