Dynamic Contrast-Enhanced (DCE) MRI is widely used for assessing changes in diseased tissue throughout the body. In DCE-MRI, a contrast agent is injected into the bloodstream. Quantitative measurements of various parameters are made by monitoring the concentration of the contrast agent as it passes through a region of interest (ROI). Estimation of the parameters requires a specified arterial input function (AIF). In many cases the time course, TC, of the concentration of contrast agent in the ROI can be represented by a simple two compartment model described by three parameters. In many imaging cases either no suitable artery exists in the imaging field of view, or the AIF is inaccurate. This work uses an iterative method for blind estimation to determine quantitative parameters without a priori knowledge of the AIF.

The MRI studies were performed on a 1.5T system using a T1-weighted spoiled gradient echo sequence. The TCs were partitioned into clusters using a k-means algorithm. The average TC from each cluster was used as input in the blind estimation. The algorithm alternately varied the compartmental parameters and the estimated input function to obtain a solution. During each iteration, the estimated blood curve was then fit to a particular analytical form for the AIF consisting of two gamma variate curves and a sigmoid function. This form was found to closely match the sharp rise in concentration when the contrast bolus first reaches the ROI, as well as the second pass of this bolus and subsequent washout of concentration. This particular form for the AIF requires 11 parameters to be fully described. Fitting the AIF to this form reduced noise amplification inherent in the blind estimation process.

The blind estimation process was applied to twelve sets of DCE-MRI patient data for which it was possible to determine measured input functions from arteries in the field of view. The estimated AIF closely matched the measured AIF when arterial voxels were included in the estimation. When arterial voxels were excluded from the estimation, the resulting AIF was dispersed in time as compared to the measured AIF. In situations with particularly noisy AIFs, or when no easily measured AIF is available, blind estimation provides an accurate method for quantification of DCE-MRI.