Linear dynamic modelling and Bayesian forecasting of tumor evolution

We consider a linear dynamic model for tumor growth evolution. A number of temporal statistical models for tumor growth exist in the literature. In the majority of these cases the employed models are formulated in a deterministic context, providing no information on their uncertainty. Some of these are theoretically well defined and very useful in practice, e.g. to define general optimal treatment protocols through nonlinear constrained optimization. Nevertheless a challenging task is the estimation of the model parameters for a specific individual since, especially in humans, it is not feasible to collect a large number of tumor size values with respect to time, as the tumor is removed immediately after diagnosis in most cases. Therefore, we suggest a probabilistic model for personalized sequential tumor growth prediction, given only a few observed data and an a priori information regarding the average response to a specific type of cancer of the population to which the subject belongs. We validated the proposed model with experimental data from mice and the results are promising.