Metamodeling for ultra-fast parameter estimation : Theory and evaluation of use in real-time diagnosis of diffuse liver disease

Abstract: Diffuse liver disease is a growing problem and a major cause of death worldwide. In the final stages the treatment often involves liver resection or transplant and in deciding what course of action is to be taken it is crucial to have a correct assessment of the function of the liver. The current “gold standard” for this assessment is to take a liver biopsy which has a number of disadvantages. As an alternative, a method involving magnetic resonance imaging and mechanistic modeling of the liver has been developed at Linköping University. One of the obstacles for this method to overcome in order to reach clinical implementation is the speed of the parameter estimation. In this project the methodology of metamodeling is tested as a possible solution to this speed problem. Metamodeling involve making models of models using extensive model simulations and mathematical tools. With the use of regression methods, clustering algorithms, and optimization, different methods for parameter estimation have been evaluated. The results show that several, but not all, of the parameters could be accurately estimated using metamodeling and that metamodeling could be a highly useful tool when modeling biological systems. With further development, metamodeling could bring this non-invasive method for estimation of liver function a major step closer to application in the clinic.