Dynamic Contrast Enhanced MRI of Carotid Plaque: Comparison of Pharmacokinetic Models

Objective The rupture of an atherosclerotic plaque is a major cause of ischemic events, such as stroke and myocardial infarction. Plaque vulnerability has been linked to the formation of angiogenic microvessels within the plaque. It has been shown that dynamic-contrast enhanced MRI (DCE-MRI) allows the detection and quantification of this neovascularisation. Pharmacokinetic models are used to describe and quantify the tissue enhancement in the plaque after injection of a contrast agent. However, the accuracy of the determined pharmacokinetic parameters is highly dependent on the chosen model. In the present study different models were evaluated with regard to their ability to describe DCE-MRI of carotid plaques. Materials and Methods Three pharmacokinetic models (Patlak (a), Extended Tofts (b), Tofts (c)) were compared in a group of 29 patients with moderate (50-69%) carotid stenosis. The differential equation of the two-compartment model and the three model solutions are given on the right. C is the concentration in plasma (p) and total plaque (tot), v is the fractional volume of plasma (p) and extracellular extravascular space (e). K is the (transendothelial) volume transfer constant [min], which indicates both blood supply and permeability. Imaging was performed on a 1.5 T MRI scanner (Intera 1.5 T, Philips Healthcare, Eindhoven, The Netherlands) with a unilateral 47 mm diameter surface RF coil placed at the position of the carotid bifurcation. Dynamic images were acquired for 11 overcontiguous slices of 6 mm thickness, using a 3D spoiled gradient echo sequence with cardiac gating (TE=3ms, TR=12 ms, α=35 ̊, 256 matrix, FOV=100 x 100 mm). Injection of the contrast agent (0.1 mmol/kg body weight of gadopentate dimeglumine) was started during the third dynamic scan using a power injector set to an injection rate of 0.5 ml/sec. Regions of interest (ROIs) in the plaque were drawn on a pre-contrast anatomic image and signal intensity curves were evaluated on a ROI-averaged basis. Analysis was performed using a standardized arterial input function (AIF) that was derived from three high temporal resolution scans and shifted to coincide with the injection of the contrast agent. Comparisons of the overall fit errors and the uncertainties for the pharmacokinetic parameters were performed to determine the best model. Furthermore, the effect of measurement time on the parameter estimation of the Patlak model was investigated using Monte Carlo simulations. For this, a model tissue residue function and the standard deviation of noise were determined from experimental datasets of 16 patients and the same AIF as in the analysis of the patient data was used. Results Typical enhancement curves for the three models are shown in Figure 1. Since the vascular term is neglected in the Tofts model, it cannot account for the first pass peak of the contrast agent. Consequently, the Tofts model had a significantly higher fit error than the other two models (Fig. 2). Fit uncertainties for K were 20 ± 2 %, 35 ± 4 % and 10 ± 1 % (mean ± SE) for Tofts, Extended Tofts and Patlak, respectively. For νe the uncertainties were 26 ± 3 % and 200 ± 53 % for Tofts and Extended Tofts, respectively. Uncertainty in νp was 67 ± 20 % for Extended Tofts and 24 ± 3 % for Patlak. The simulation results (Fig. 3) show a decrease of the estimated K values with measurement time for the Patlak model, thus indicating a bias in the estimation of K. Discussion Because of the relatively short measurement time little or no reflux can be seen in the 0 1 2 3 4 5 6 7 8 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2