Synopsis

The parameters estimated in DCE-MRI studies vary greatly depending on the arterial input function. Moreover, a more complex model than the extended Tofts’ model (ETM) is needed in glioma patients. In this work, a method to generate a voxel-specific input function (VIF) is introduced and used with a two-tissue compartment model (2TCM) that separates the fast and slow kinetics of the Gd contrast agent. The VIF provided more accurate results in the superior sagittal sinus (SSS) than the internal carotid artery and combined with the 2TCM, significantly improved the fits to the tumour over the ETM using the SSS.

Purpose

Methods

Ten patients with glioma (three grade II, three grade III and four grade IV) underwent a pre-surgical DCE-MRI examination on a 1.5T Philips Achieva MR scanner with a SENSE coil. A dual-bolus protocol⁴ was adapted by acquiring the pre-bolus sagittal scan at twice higher temporal resolution than the main-bolus transaxial scan². Pharmacokinetic analysis was performed at the voxel level using the ETM and 2TCM. The latter model assumes that the Gd contrast agent is exchanged between plasma and a fast (K_1f,k_2f) or a slow (K_1s,k_2s) tissue compartment in parallel, with the tissue concentration given by:

$$C_t(t)=(K_{1f}·e^{-k_{2f}t}+K_{1s}·^{-k_{2s}t})·C_A(t)+v_p·C_A(t)$$

where v_p is the fractional plasma volume, C_A(t) is the AIF and the bolus arrival time is included as an additional fitted parameter. The aggregated parameters K_1T and k_2T can be estimated as a function of the four rate constants, weighted by the fraction of the corresponding extravascular spaces (V_T=K₁/k₂). Three AIFs were used for the pharmacokinetic analysis: i) for the AIF_ICA, the peak was extracted from the ICA (pre-bolus) and combined with the tail of the SSS (main bolus)⁴, ii) the AIF_SSS was extracted from the SSS (main bolus), and iii) the VIF was generated under the assumption that the AIF_SSS is equal to the AIF_ICA convolved with a dispersion function D(t) when n=1 using:

$$FFT\{VIF(t)\}=FFT\{AIF_{ICA}(t)\}·FFT\{D(t)\}^n$$

where the power parameter n increases with the magnitude of the dispersion of the AIF_ICA and was included as an additional fitted parameter constrained to between 0 and 2 (figure 1).

Parametric maps were generated after correction for the non-uniform noise distribution². The tumour region was manually delineated using the post-contrast T₁-weighted and FLAIR images. Currently, two patients with glioblastoma have been fully analysed and their results reported here.

Results

With both compartmental models, the AIF_ICA provided unrealistically high K^trans and K_1f in the SSS while v_p was much lower than unity (0.39 and 0.63 for patients 1 and 2, respectively, figure 2). The use of the AIF_SSS or the VIF reduced K^trans and K_1f by half or more with either model, and increased v_p to 0.77-0.99 for both patients.

The two models exhibited different distributions of the power parameter n when the VIF was used (figure 3). In the enhancing rim of the tumour, the ETM returned median n values above unity and even close to the maximum of 2 in patient 2 whereas the 2TCM exhibited a wide distribution across the allowed values, as expected, with median n values below unity for both patients.

Estimates of v_p from either compartmental model was systematically lower when the AIF_ICA was used (0.028-0.046) compared to the AIF_SSS (0.040-0.070) and the VIF (0.037-0.083), which showed similar distributions (figure 3). For both K^trans and K_1T, the AIF_ICA yielded higher values (0.045-0.070 ml_p/min/ml_t) than the other two input functions (0.028-0.039 ml_p/min/ml_t), in agreement with previous findings¹. K^trans estimates from the ETM using the AIF_SSS were also higher than those using the VIF whereas K_1T estimates from the 2TCM were similar using these two input functions. The distributions for v_e and k_2T using the different input functions were more similar than for the other parameters.

The 2TCM described the data significantly better than the ETM in the enhancing rim of the tumour (F-test, figure 4). Moreover, fits to areas of the tumour were significantly improved by using the VIF instead of the AIF_SSS with the 2TCM. Parametric maps from either the ETM with the AIF_SSS or the 2TCM with the VIF exhibited heterogeneity in the tumour for all parameters (figure 5).

Conclusion

Acknowledgements

References

1. Keil C.V, Mädler B, Gieseke J, et al. Effects of arterial input function selection on kinetic parameters in brain dynamic contrast-enhanced MRI. Magn. Reson. Med. 2017; 40:83-90

2. Krokos G, Thacker N, Djoukhadar I, et al. Non-uniform noise correction in dynamic contrast-enhanced MR images reveals superiority of the two compartmental exchange model over the extended Tofts and the adiabatic approximation to the tissue homogeneity models in glioma patients. Proc. Intl. Soc. Mag. Reson. Med. 24; 2016; abstract: 2465

3. Lüdemann L, Grieger W, Wurm R, et al. Comparison of dynamic contrast-enhanced MRI with WHO tumor grading for gliomas. Eur. Radiol. 2001; 11(7):1231-41

4. Li KL, Buonaccorsi G, Thompson G, et al. An improved coverage and spatial resolution-using dual injection dynamic contrast-enhanced (ICE-DICE) MRI: a novel dynamic contrast-enhanced technique for cerebral tumors. Magn. Reson. Imag. 2012; 68:452-62