Data Analysis and Statistical Methods

All calculated maps (different CBVs, 𝐾𝐾₂’s) were once analyzed voxel-based within each patient, directly comparing the results of different correction methods and the effects of correction. Secondly, with respect to VOIs, values were averaged across tumor and healthy tissue regions and analyzed for all patients. Furthermore, the median and standard deviations within the VOIs were considered. In order to characterize the curve shapes in particular for comparisons of different acquisitions, rPH and rPSR values were also evaluated.

Sequence design: The impact of sequence parameters and designs (FA 70° versus FA 90°

and PRESTO versus 2D EPI) on correction methods was examined in relation to simulations.

Comparison of correction methods: The combination of different procedures and acquisition schemes (only pre-bolus, only post-processing, pre-bolus and post-processing) applied to data from the hypoxia study was used to evaluate the most reliable and stable correction method. The amount and direction of corrections were investigated and compared with each other, and differences and agreements of the methods were evaluated. Since for reliable results, an accurate fitting is one of the basic prerequisites, the fitting residuals and individual curves were inspected carefully. Further, the impact of integration intervals (full versus first pass integration) on these correction methods was examined (section 3.7.4). Though both intervals are equally used in the literature [55], in leakage affected data a direct comparison has not yet been performed. Thus, the effect on absolute and normalized CBV values in presence of CA leakage was investigated.

The eight patients, where two DSC scans were acquired in one session, were used to evaluate the correction methods. The first dose served as a pre-bolus to minimize T1 leakage effects in the second acquisition (section 3.2.1). In case of appropriate post-processing, the significance of this technique should be reduced. The hypothesis that adequate post-processing would homogenize CBV results between the first and the second bolus, should allow figuring out the most reliable and stable correction methods.

Likewise, using the accordance of results from both boli as an indicator for processing quality, further investigations concerning the integration range, the relationship between MR signal and CA dose and timing corrections were done.

Evaluation of Methods in Patient Data

Impact of contrast agent: All correction methods were further investigated using DSC-MRI acquired with different contrast agents (Gd-DTPA, gadofosveset trisodium). The more intravascular contrast agent gadofosveset trisodium (Vasovist) is generally used for angiography. Its usefulness for perfusion and extravasation imaging has not been investigated so far. Vasovist does not reenter (reflux = 0) the vasculature once extravasated.

In case of disrupted BBB, it should better fit model assumptions for CBV leakage correction methods. Further, its relaxivity is higher and thus, a lower contrast agent dose should be necessary. Analyzed were all different leakage correction methods, ATCs and integration ranges. Here, the accordance between CBV results obtained with both contrast agents was used as an indicator for the validity of CBV estimates. Furthermore, results were used to better assess the meaning of 𝐾𝐾₂ values.

Validation of CBV^DSC with DCE and PET: DCE-MRI and PET provided independent reference values for CBV (CBV^DCE and CBV^PET). Theoretically, both methods allow a more accurate conversion of the measured signal into CA concentration.

With PET, it is possible to create images of absolute CA concentrations (section 2.2).

DCE is based on T1 enhancement, and thus the linear relationship between concentrations and ∆𝑅𝑅1 is valid for a wider value range (section 2.5). Since the spatial resolution of both techniques is lower than in DSC, for voxel-wise comparisons CBVDSC

was smoothed with a Gaussian kernel (5 × 5 × 5 voxels). Since modeled CBV^PET values are only reliable within the area of [¹⁸F]FET accumulation, these comparisons were focused on VOI^TUMOR.

Potential usefulness of permeability related values: An additional output of all DSC-based leakage correction methods is the parameter 𝐾𝐾₂, that was used to study the heterogeneity of tumors. This was done by calculating the percentage of the tumor volume (VOI^CET) showing predominant T1 effects (𝐾𝐾₂ > 0). In detail, the distribution of both effects and their extent were studied. Additionally, the influence of the PB was investigated and two different contrast agents were compared against each other. The relation between 𝐾𝐾₂ and vessel permeability was investigated using the 𝐾𝐾^{𝑡𝑡𝑙𝑙𝑝𝑝𝑢𝑢𝑡𝑡} values.

Since DCE-MRI is the method of choice for permeability imaging, 𝐾𝐾^{𝑡𝑡𝑙𝑙𝑝𝑝𝑢𝑢𝑡𝑡} was used as a reference, even though in these patients only DSC data acquired after a pre-bolus were available, which potentially confounds the analysis.

Impact of CBV variability on rOEF: In a numerical study, the impact of altered CBV values on rOEF estimation was evaluated using Eq. (3.23) inserting realistic CBV deviations found in this study (-100 % to +200 %). In healthy brain tissue, a typical deoxygenated CBV of 3 ml/100 g, which corresponds to a total CBV of approximately 4 ml/100 g [110], can be assumed and therefore was used as reference CBV. Considering further a homogeneous OEF of about 0.4 in healthy brain tissue [111], at 3.0 T 𝑅𝑅2′ has to be 3.8 s^-1. Using Eq. (3.23) the change of rOEF according to the CBV variations due to different analyses methods, as observed in patient data, could be estimated. Further,

the quality of rOEF maps, calculated with four different CBVs (CBVunc 1 and CBVmethod I, each with and without pre-bolus acquisition) was visually inspected for two exemplary patients.

Statistical tests: The Wilcoxon signed rank test was used for statistical test regarding the comparisons between parameter maps of the same patient as well as for comparisons within the same patient cohort. For comparisons between different patient groups (hypoxia group versus double bolus group), the Wilcoxon-Mann-Whitney test was used.

The consistency of different methods was described by the reproducibility coefficient (RPC = 1.96 standard deviations). A small RPC corresponded to high reproducibility or accordance of methods. Boxplots were used for group comparisons. VOI- and voxel-wise correlations were done with Pearson (CBV) and Spearman (𝐾𝐾₂) correlation, unless stated otherwise. The correlation coefficient 𝑟𝑟 thereby served as quality parameter.

4 Results

Some preliminary results of an initial method comparison have already been published [63]. The following results are based on more extended simulations as well as evaluations of a larger and an additional patient group.

4.1 Simulations

Simulations were performed to investigate the influence of image quality, physiological variations and processing parameters on the performance of individual extravasation correction methods and the resulting CBVs. The focus was thereby primarily on the individual dependence of methods on different parameters and their relation to other methods. Prior to this, the preferred type of the arrival time correction (ATC) and the optimal definition of the extravasation phase for method III variants have been investigated with regard to more accurate CBV values. The basis for all CBV calculations were perfusion curves simulated without and with contrast agent extravasation (section 3.6).