Parametric modeling of DSC-MRI data with stochastic filtration and optimalinput design versus non-parametric modeling
Abstract In the paper MRI measurements are used for assessment of brain tissueperfusion and other features and functions of the brain (cerebral blood flow - CBF, cerebral bloodvolume - CBV, mean transit time - MTT). Perfusion is an important indicator of tissue viability andfunctioning as in pathological tissue blood flow, vascular and tissue structure are altered withrespect to normal tissue. MRI enables diagnosing diseases at an early stage of their course. Theparametric and non-parametric approaches to the identification of MRI models are presented andcompared. The non-parametric modeling adopts gamma variate functions. The parametricthree-compartmental catenary model, based on the general kinetic model, is also proposed. Theparameters of the models are estimated on the basis of experimental data. The goodness of fit of thegamma variate and the three-compartmental models to the data and the accuracy of the parameterestimates are compared. Kalman filtering, smoothing the measurements, was adopted toimprove theestimate accuracy of the parametric model. Parametric modeling gives a better fit and betterparameter estimates than non-parametric and allows an insight into the functioning of the system. Toimprove the accuracy optimal experiment design related to the input signal was performed.
Keywords Magnetic Resonance Imaging; Models, Theoretical; Statistics, Nonparametric
Annals of Biomedical Engineering
0090-6964, Volume 35, Issue 3, 2007, Pages 3-464
Kalicka,R; Pietrenko-Dabrowska,A
