acquisition protocols, the pharmacokinetic modeling and the types of contrast agent, however, are very diverse. Their use seems more often driven by the circumstances than chosen on a rational basis. We have tested on the same data sets several pharmacokinetic models to find a rationale for using these models. Materials and Methods: We have tested several DCE-MRI and DCE-CT acquisitions obtained on different types of tumors with a high temporal resolution (1 image per 1-3 s) and during a long acquisition time (10 min when using extracellular contrast media and 30 min when using macromolecular contrast media). From these data we, have generated several sets of data by varying the acquisition time and the temporal resolution (by shortening the acquisition and/or by deleting intermediate images). Finally, we have tested different types of compartmental models: the comprehensive model (taking into account all the volumes and all the exchanges between compartments), a threeparameter model neglecting the return leak, the Patlak model, the Tofts model and the extended Kety model, all using the measured arterial input function (AIF). The quality of fit was tested using the FIM criteria; the redundancy and the precision were tested using the covariance matrix.Results: First we have demonstrated, using the same type of mathematical notation, that all the models derive from the same comprehensive model. On the data, it appeared that, when a high temporal resolution and a sufficiently long acquisition time were used, the comprehensive model was the most appropriate, but necessitated a sufficient SNR. The comprehensive model gives simultaneous access to the tissue perfusion (FT), the fractional blood volume (VB), the permeability surface product (PS), the fractional interstitial volume (Ve) and the artery to tissue lag time (Dt). With a shorter acquisition time and a lower temporal resolution, the Patlak model appeared adequate, but yielding only VB and PS. With a longer acquisition time the extended Kety (or extended Tofts) model was required, yielding to VB, PS and Ve. The simple Tofts model gave the worst results due to the fact that this model neglects the large VB found in tumors. Conclusion: We have introduced a unified theory based on the comprehensive model that connects the different compartmental models used for DCE imaging. We will expose a rational strategy to choose the most adequate model for acquired data, and to design acquisition protocols in accordance with the expected microvascular parameters and the type of contrast agent. Francisco, CA, USA). Kinetic analysis of the dynamic enhancement responses based on a uni-directional model were used to estimate fractional plasma volumes (fPV) and the coefficients of endothelial permeability (K PS ). Vascular richness, estimated by MRI as the fPV (%), was correlated in the same tumors with the quantitative fluorescent microscopic assay of the density (%) for the perfused-vascular bed based on lectin staining. The same quantitative MRI assays, enhanced with a biotin-l...
