Data-enhanced prediction of organ-specific tumor growth in the liver - a hybrid knowledge and data-driven approach

We aim to enable near real time computation of the high-fidelity multiphase and multiscale simulation of tumor growth in human liver via data-driven surrogate models. This will allow patient-specific prognosis in clinical environments. We will design hybrid data- and knowledge-driven surrogate models for ODE, PDE, and coupled PDE-ODE models. To improve the coupling performance, we will focus on efficient coupling using preCICE. Finally, we will enable individualization of the simulation by developing a pathway to incorporate relevant patient data (individual tumor volume, preconditions or blood pressure) in close collaboration with clinical partners from Jena University Hospital. Patient-specific model parameters are generally subject to aleatory uncertainty. To estimate the confidence interval in the prognosis via the surrogates, Bayesian neural networks (BNNs) will be applied.