Model-driven registration for multi-parametric renal MRI

The use of MR imaging biomarkers is a promising technique that may assist towards faster prognosis and more accurate diagnosis of diseases like diabetic kidney disease (DKD). The quantification of MR Imaging renal biomarkers from multiparametric MRI is a process that requires a physiological model to be fitted on the data. This process can provide accurate estimates only under the assumption that there is pixelto-pixel correspondence between images acquired over different time points. However, this is rarely the case due to motion artifacts (breathing, involuntary muscle relaxation) introduced during the acquisition. Hence, it is of vital importance for a biomarkers quantification pipeline to include a motion correctionstep in order to properly align the images and enable a more accurate parameter estimation. This study aims in testing whether a Model Driven Registration (MDR), which integrates physiological models in the registration process itself, can serve as a universal solution for the registration of multiparametric renal MRI. MDR is compared with a state-of-the-art model-free motion correction approach for multiparametric MRI, that minimizes a Principal Components Analysis based metric, performing a groupwise registration. The results of the two methods are compared on T1, DTI and DCE-MRI data for a small cohort of 10 DKD patients, obtained from BEAt-DKD project’s digital database. The majority of the evaluation metrics used to compare the two methods indicated that MDR achieved better registration results, while requiring significantly lower computational times. In conclusion, MDR could be considered as the method of choice for motion correction of multiparametric quantitative renal MRI.