Iterative computing of decoupled interacting models

Abstract

Physiological processes in the human body as well as their reaction to changes in the therapy regime can be predicted by mathematical models. Medical Decision Support Systems (MDSS) can employ these predictions in order to optimize therapy settings. When optimizing ventilation therapy in critically ill patients, these predictions should be extended to also consider other parts of the human body. A previously presented framework allows combination of three model families (respiratory mechanics, cardiovascular dynamics and gas exchange) to provide a broader picture when predicting the outcome of a therapy setting. The three model families are combined to form a complex model system with interaction submodels.In the previously presented framework, all submodels were computed as a tightly coupled system, i.e. all models are evaluated at every simulation time step. Tests showed that this is computationally very costly. Therefore, a decoupled computing approach is proposed. Direct model interaction is not possible with this approach. Therefore, computing is done iteratively, where results of the preceding iteration are used as interface signals in the actual iteration.Results show that simulation error stays constant after a maximum of three iterations. Maximum simulation error showed to be 1.4% compared to the previously proposed coupled computing approach. Simulation time was decreased by factor 56 using one iteration and factor 26 using two iterations.KeywordsPhysiological modelinginteracting modelsdecoupled computingiterative evaluationdecision support