A Bayesian framework for pharmacokinetic modelling in dynamic contrast-enhanced magnetic resonance cancer imaging

Dynamic contrast-enhanced resonance imaging (DCE-MRI) plays an important role in oncology. During a number of scans, a contrast agent, usually a Gadolinium complex like gadopentate (Gd-DTPA) is injected into the patient. DCE-MRI scans show the flow of the contrast agent, and therefore the blood flow, between vascular space and extracellular extravascular space (EES), since the contrast agent is too large to enter the cells. Growth of tumor depends on its ability to initiate formation of new blood vessels, that can grow into the tumor; a process called angiogenesis. So tumors are regions of high blood flow and of high fraction of vascular space, and therefore can be detected via DCE-MRI. We use standard pharmacokinetical models for DCE-MRI as basis for our data model in a Bayesian hierarchical framework. After describing the standard procedure in DCE-MRI, we develop Bayesian hierarchical model to estimate the kinetic parameters. In a further step we include spatial information via a Gaussian Markov random field (GMRF) prior. Cancer tissue often is heterogeneous; therefore smoothing techniques for DCE-MRI need to have edgepreserving qualities. Also, coil effects and other sources of errors differ over the field of view. Therefore we use an adaptive GMRF approach, which estimates local smoothing weights.