Use of surrogate models in tumor growth modeling

Cancer modeling is a class of diseases that span mechanisms occuring in different time and space scales. Current modeling framework combines continuum models and individual based models to better represent such heterogeneous multiscale dynamics. Due to inherent uncertainties involved in this kind of problem, Bayesian inference for parameters estimation [1] is an adequate methodology since it allows taking into account uncertanties due to the presence of error in the measurements as well as model inadequacies. It relies on simulating the forward model for many possible different configurations of the parameter set, which can lead to an overwhelming computational burden. Surrogate models have been developed in the literature to alleviate this issue. According to the source, this approach can be also named as a metamodel, reduced model, response surface, among others. The general idea is to build surrogates for the quantity of interest that can be cheaply and accurately evaluated without requiring to run expensive models. The accuracy of such strategy for Bayesian inference is still an open issue [3]. [...]