However, MI-773 lacking of accurate tubule parameters estimation was supported and underestimated GFR was found. Later, a similar model was employed to measure GFR in patients by using the whole kidney ROI, which was just simplified by Etofylline ignoring the dispersion effect in the original two-compartment model. Then, a separable compartment model, of which the time delay of contrast was assumed to be zero, provided estimated parameters which were analogues to those of the original two-compartment model. Besides these models, three-compartment models were also proposed for GFR estimations. Nevertheless, the variability of GFR measurement was more pronounced than that of twocompartment models. In addition, the assessment of GFR was sensitive to the segmentation of cortex and medulla, which may result in an unstable measurement. The increase of compartment number would result in an expense of system complexity, and a simple cortical compartment model is alternative because the glomerular filtration occurs in the cortex. We borrowed Zhang et al��s idea of using impulse residue function to improve the robustness in GFR measurements, and combine it into a two-compartment model. In this study, a modified two-compartment model was proposed by introducing the impulse residue function to obtain effective estimations of GFR and renal plasma flow from DCE-MRI. The advantage of the new model in GFR or RPF measurements over other published models was investigated in Monte Carlo simulation under different noise levels. Then, quantitative estimations were performed in healthy rabbits by using our new model. Furthermore, the proposed model was employed to measure GFR changes in rabbits with unilateral ischaemic acute kidney injury to test its validity. Pixel-wise calculation was performed and the cortical GFR results were compared with its contralateral kidney. The major focus of this study was put on the sufficient robustness in kinetic parameters estimation and the ability to discriminate healthy and diseased kidneys. Renal parenchyma and the aorta were automatically segmented from the surrounding tissue with a Level-Set framework.