[home] [Personal Program] [Help]

---

Session: Medical Instruments - Surgery II
Session starts: Friday 25 January, 13:00
Presentation starts: 13:00
Room: Lecture room 557

---

Jörn Kretschmer (Institute of Technical Medicine)
Knut Möller (Institute of Technical Medicine)

Abstract:
Mathematical models are a major tool in providing insights in the physiological processes of the human body. They can be exploited to predict effects of therapeutic strategies in intensive care medicine. Medical Decision Support Systems (MDSS) might capitalize such predictions when searching for the optimal therapeutic setting. In critically ill patients that e.g. depend on mechanical ventilation these predictions should cover all main involved organ systems, such as pulmonary mechanics, gas exchange and cardiovascular dynamics. In a previously presented framework developed to support mechanical ventilation we combine elements of these three organ systems [1]. A complex interacting model system can be formed arbitrarily from submodels of these model families. Interaction is achieved through interfaces that enable the submodels to exchange parameters. Computing combinations of moderately complex submodels showed to be computationally costly. An MDSS needs to evaluate the model systems numerous times to find the optimal therapeutic setting. Thus, high computation costs would delay the treatment recommendation. Decoupling the interacting submodels allows for individual computation of submodels with different system dynamics. Unfortunately, direct model interaction as implemented in a coupled computing approach is not possible when submodels are decoupled. Therefore, interface signals need to be substituted by estimates. These estimates can be calculated by the decoupled models and can be improved by iterating the computation. Although it might seem that computing costs rise when iterating the simulation, the application of individual solvers should compensate the increased number of calculations needed. A test setup was created containing a respiratory mechanics model of third order, a tidal breathing 4-compartment gas exchange model [2, 3] and a 19-compartment cardiovascular model which is reactive to intrathoracic pressure [4]. Simulation error converged to a minimum after three iterations. Maximum simulation error after three iterations showed to be 1.1% compared to a coupled computing approach. Simulation error was found to be below measurement noise generally found in clinical data. Simulation time was reduced by factor 17 using three iterations. Applying the proposed calculation scheme, moderately complex model combinations can be made applicable for model based decision support.