New Hybrid Technique for Accurate and Reproducible Quantitation of Dynamic Contrast-Enhanced MRI Data Ka-Loh Li 1 and Alan Jackson 2 * The accuracy and precision of two major approaches for ana- lyzing dynamic MRI data from the first passage of a Gd-DTPA contrast bolus are examined, using a Monte Carlo simulation. Method 1 fits the contrast concentration curve of the first pass to a two-compartment kinetic model to determine the tissue pharmacokinetic parameters. Method 2 decomposes intravas- cular and interstitial components of the first-pass curve based on a leakage profile (LP) model. Based on the results of these Monte Carlo simulations, a new “hybrid” method is proposed that combines both analytical approaches to optimize accuracy and precision of estimates of K trans , v e , and v p . The new method was evaluated by computer simulation and used on experimen- tal results from a patient with primary brain tumors. The new method has the potential to provide more accurate quantifica- tion of tissue plasma volume and vessel permeability. Magn Reson Med 50:1286 –1295, 2003. © 2003 Wiley-Liss, Inc. Key words: dynamic MRI; blood volume; endothelial permeabil- ity; Monte Carlo simulation The improved temporal resolution of dynamic MRI se- quences means that short-lived variations in signal inten- sity (SI), which occur during the first passage of a bolus of contrast agent (CA), can now be accurately characterized. This has led to increasing interest in the development of data analysis methods for extracting pharmacokinetic pa- rameters, such as the volume transfer constant (K trans ) between plasma and the extracellular/extravascular space (EES), fractional volume of EES (v e ), and fractional plasma volume (v p ) from first-pass DCE-MRI data (1–5). Conventionally, investigators have fit pharmacokinetic models of CA distribution to CA concentration curves using a variety of curve-fitting techniques (6 –9). However, several groups have identified a potential lack of stability in parameter estimation when multiparametric curve-fit- ting (MPCF) techniques are used (1,10 –14). Local minima may cause failures to converge to appropriate minimal values (13), and the possibility of a large covariance be- tween the fitted parameters may cause different combina- tions of fitting variables, resulting in competing “best fit” solutions. These errors occur especially when signal-to- noise ratio (SNR) is low, which is particularly common when pixel-by-pixel parametric mapping is performed. We previously described an alternative approach, the first-pass leakage profile (FPLP) method (1,2), which de- composes the intravascular and interstitial components of the tissue residue function during the first pass of the CA. Our previous studies in patients with cerebral glioma and hepatic tumors (1,2,15) demonstrated that this method is highly reproducible. However, it is based on the assump- tion that backflow of the CA from interstitial water to plasma is negligible during the first passage of CA, which may be expected to adversely affect the accuracy of K trans estimates, particularly where CA extraction fractions are high. Only a small number of publications have addressed the issue of accuracy and reproducibility of calculated phar- macokinetic parameters such as perfusion (16), blood vol- ume (17), and capillary permeability surface area product (14,18 –22). The accuracy and precision of these estimates depends on the accuracy of the models themselves, and on the investigator’s ability to separately identify the second- ary effects of variations of individual model parameters in real data sets (23). This is affected not only by the param- eter estimation method, but also by the sampling fre- quency, sampling period, signal-to-noise ratio (SNR), and the true values of the physiological parameters. In this study we present 1) a Monte Carlo simulation to assess the accuracy and precision of the MPCF and FPLP methods in analyzing first-pass T 1w DCE-MRI data; 2) a new “hybrid” analysis method designed to more accurately quantify the changes of plasma volume and vessel permeability in tis- sues based on the findings of the Monte Carlo simulation; 3) a Monte Carlo simulation to assess the accuracy and precision of the “hybrid” analysis technique; and 4) an example of the new method applied to data from a patient with primary brain tumors. METHODS The terminology used in this work follows the conven- tions described by Tofts et al. (24). Monte Carlo Simulations for Error Analysis Simulation of Vascular Input Function The Monte Carlo simulations used a simulated vascular input function (VIF) derived from data collected in a series of 10 patients with cerebral glioma. Details regarding the patients and image acquisition sequence have been de- scribed fully elsewhere (1,25). Measurements of the tracer concentration in blood plasma (C p (t)) were obtained from voxels in the vertical part of the superior sagittal sinus (SSS) in all 10 patients (26,27). The simulated vascular input function was constructed by fitting the average C p (t) from all 10 patients using a gamma function concatenated with a steady-state concentration (28). 1 Department of Radiology, University of California–San Francisco, San Fran- cisco, California. 2 Division of Imaging Science and Biomedical Engineering, Stopford Medical School, Manchester University, Manchester, UK. *Correspondence to: Professor Alan Jackson, Division of Imaging Science and Biomedical Engineering, Stopford Medical School, Manchester Univer- sity, Manchester M13 9PT, UK. E-mail: Alan.Jackson@man.ac.uk Received 20 September 2002; revised 13 March 2003; accepted 14 March 2003. DOI 10.1002/mrm.10652 Published online in Wiley InterScience (www.interscience.wiley.com). Magnetic Resonance in Medicine 50:1286 –1295 (2003) © 2003 Wiley-Liss, Inc. 1286