To test the utility of a modified method for correction of contrast agent extravasation in DSC-MRI which is independent of tumor mean transit time (MTT) based on parametric analysis of the tissue impulse response function combined with automated global arterial input function (AIF) approximation. Contrast agent extravasation is known to be a serious confounder in estimations of cerebral blood volume (CBV) analysis of primary brain tumors using DSC-MRI (1). It has been recently shown that leakage correction based on analysis of the tumor impulse response function (IRF) is less sensitive to mean transit time (MTT) -induced error compared to an established method as originally proposed by Weisskoff et al. (2). The usefulness of the MTT-insensitive approach reported in (3) is however limited by the need to determine the AIF in order to obtain the IRF and further by the empirical determination of the leakage-dependent portion of the IRF. Here, we propose a modified method whereby the AIF is estimated directly from the mean brain tissue response and the contrast agent leakage is derived from a full parametric curve fit of the IRF using the two-compartment uptake model.

Kinetic model: The two-compartment uptake model (4) incorporates the effect of finite MTT and extravasation but assumes absence of significant contrast agent reflux (K^trans/ v_e ≈ 0). The IRF is then given by:

$$I(t)=exp(-t/T_{p}) + E\cdot(1-exp(-t/T_{p}))$$ (Eq.1)

where E is the extraction fraction = K^trans/(F+K^trans), F is tissue flow and MTT= T_p/(1-E). Leakage corrected CBV is then given by :

$$CBV_{corr}= F\cdot T_{p}/(1-E) $$ (Eq.2)

AIF estimation: An idealized AIF was estimated from the measured global tissue response (excluding tumor tissue) by assuming an exponential IRF in (non-leaky) brain tissue so that: $$\overline{C_{t}(t)}=C_{a}(t)\otimes \overline{F} \cdot exp(-t/\overline{MTT})$$ (Eq. 3)

where $$$\otimes$$$ is the convolution operator, $$$\overline{C_{t}(t)}$$$ is global tissue response, $$$ \overline{F}$$$ is global mean perfusion, $$$\overline{MTT}$$$ is the global mean transit time and C_a(t) is the estimated AIF. C_a(t) can then be determined by SVD deconvolution by assuming fixed values for $$$ \overline{F}$$$ and $$$\overline{MTT}$$$.

Twenty eight patients with recurrent glioblastomas enrolled in a phase II clinical trial of the VEGFR inhibitor cediranib (clinicaltrials.gov, NTC00305656) were imaged twice on separate days prior to cediranib administration. Imaging was performed at 3 T (Magnetom Trio, Siemens Medical) and the protocol included DSC-MRI, pre-and post T1- weighted spin-echo images used for expert definition of tumor regions-of-interest (5). The DSC examination was preceded by a DCE-MRI, thereby minimizing T1-leakage effects. From the DSC data, global tissue response, excluding tumor was determined as previously described (6) and AIF was derived from Eq. 3 using $$$ \overline{F}$$$= 50 mL/100g/min and $$$\overline{MTT}$$$ = 4 sec. Using this AIF, pixel-wise estimates of F, MTT, E and K^trans were obtained by first estimating the IRF from SVD deconvolution with Tikhonov regularization (6) and then obtaining the model parameters from non-linear least squares fit of the resulting IRF to Eq. 1. The results for the proposed method (Model 1) were compared to the apparent leakage rate (K_a) and leakage corrected normalized CBV (nCBV_corr) estimated from the reference model of Weisskoff et al. (Model 2), implemented as in (3). Correlation between the estimates of K^trans/ K_a and nCBV_corr for the two models were compared using Spearman correlation coefficient (ρ) from the mean tumor values obtained from the two separate scanning sessions. Image analysis was performed using nordicICE (NordicNeuroLab, Bergen, Norway).Figure 1 shows sample cases of the generated maps using Model 1. Figure 2 shows the test-retest correlation of leakage estimated from the two models. There was no significant difference in the Spearman correlation coefficient for Model 1 versus Model 2 (McNemars test). Figure 2A shows the correlation between corrected nCBV values obtained with Model 1 versus Model 2. Although the overall correlation is high, a significant MTT-dependent difference in the nCBV ratio between the two correction methods is observed (Figure 2B), in line with the previously observed MTT-dependence of Model 2 (3).We propose a modified CBV leakage correction method for DSC-MRI applying the two-compartment uptake model and without the need for explicit AIF determination. The proposed method yields test-retest stability similar to the reference method, but with the added advantage of reduced MTT sensitivity and estimation of extraction fraction, E. The model makes global assumptions about tissue perfusion and MTT in order to construct an idealized AIF. Although average perfusion is likely to vary between individuals and as a result of pathology, the method proved to have a high test-retest stability of leakage estimations in glioblastoma patients.