Hierarchical modeling for medical decision support

Abstract

Mathematical models are widely used to simulate physiological processes in the human body. They can be exploited for diagnostic purpose or the automation of therapeutical measures, if applied appropriately. In this work a feasible, hierarchical modeling approach with associated identification processes is proposed in the context of mechanical ventilation. It allows identifying relative complex physiological model systems and to use them to support decision making on the intensive care unit (ICU). This hierarchical approach is exemplarily demonstrated with the combination of multiple hierarchical model families representing knowledge about pulmonary mechanics, gas exchange and cardiovascular dynamics. Simulation speed could be increased by a factor of almost 20 with a calculation scheme that exploits the hierarchical structure of the model system. Robust identification within this framework is illustrated by identifying the viscoelastic model (VEM) of respiratory mechanics. Evaluation with simulation data demonstrates that the hierarchical approach always revealed the correct solution while the success rate of the common direct approach exponentially decreases with decreasing quality of initial estimates. These investigations demonstrate that a hierarchical approach is beneficial with respect to flexibility, robustness and efficiency when employed in decision making at the bedside.