The Bayesian Virtual Epileptic Patient (BVEP)
The mathematical model of seizure initiation and propagation can provide predictive insight into the complex underlying mechanisms, thus, to assist the routine diagnosis of patients with pharmacologically-resistant epilepsy. However, finding the best possible prediction to the observed data is a challenging task, especially in high-dimensional nonlinear complex system, when it is required to estimate a large number of model parameters. To address this challenge in the context of epilepsy, we have developed a novel probabilistic framework, the Bayesian Virtual Epileptic Patient (BVEP), to systemically predict the location of seizure initiation and propagation in a virtual epileptic patient.
In the proposed approach, a generic mathematical model of seizure-like events (Epileptor) is integrated with patient-specific non-invasive anatomical information, and the most advanced machine-learning techniques are used to infer the spatial map of epileptogenicity across different brain regions. Using the automatic self-tuning NUTS/ADVI algorithms in probabilistic programming languages (PPLs), and appropriate form of parameterization, the BVEP accurately infers the spatial map of epileptogenicity across different brain regions, and with that, the hypothetical brain areas responsible for the seizure initiation and propagation. The convergence diagnostics and posterior behavior analysis validate the reliability of the estimations. Our results indicate that the BVEP provides a novel patient-specific strategy for epileptogenicity hypothesis testing to improve outcome after epilepsy surgery.
Full research paper:
ICEI/FENIX helps lead Virtual Epileptic Patient model to clinical trial
Find out how powerful Supercomputing on the European Fenix infrastructure, offered through the ICEI project, helped the preparation of a major clinical trial with this novel approach. Read more