On the influence of prior information evaluated by fully Bayesian criteria in a personalized virtual brain model
Abstract
Individualized anatomical information has been previously used as a prior knowledge in fully Bayesian paradigms on the validation process of virtual brain models. However, the actual sensitivity to such personalized information in priors is still unknown. The Watanabe-Akaike information criterion (WAIC) and Bayesian leave-one-out (LOO) cross-validation are two rigorous and fully Bayesian approaches for estimating pointwise out-of-sample prediction accuracy that allow us to efficiently determine the most likely among a set of hypotheses. This study describes the use of WAIC and LOO cross-validation to assess different epileptogenicity hypotheses regarding the location of pathological brain areas based on a priori knowledge from dynamical system properties. The Bayesian Virtual Epileptic Patient (BVEP) model, which relies on the fusion of structural data of individuals, is used to infer the spatial map of epileptogenicity across different brain areas. Our simulations of seizure propagation patterns demonstrate that measuring the BVEP predictive accuracy by WAIC and LOO with informative priors enables the efficient evaluation of potential hypotheses regarding the degree of epileptogenicity across different brain regions. In constrast, while using uninformative priors, the information criteria is unable to provide strong evidence about the epileptogenicity of brain areas. We also show that WAIC and LOO correctly assess different hypotheses about both structural and functional components of whole-brain network models that differ across individuals. The Bayesian approach used in this study suggests a novel patient-specific strategy for epileptogenicity hypothesis testing in generative brain network models of epilepsy to improve surgery outcome.