Quantitative MRI using relaxometry in malignant gliomas detects contrast enhancement in peritumoral oedema

Malignant gliomas are primary brain tumours with an infiltrative growth pattern, often with contrast enhancement on magnetic resonance imaging (MRI). However, it is well known that tumour infiltration extends beyond the visible contrast enhancement. The aim of this study was to investigate if there is contrast enhancement not detected visually in the peritumoral oedema of malignant gliomas by using relaxometry with synthetic MRI. 25 patients who had brain tumours with a radiological appearance of malignant glioma were prospectively included. A quantitative MR-sequence measuring longitudinal relaxation (R1), transverse relaxation (R2) and proton density (PD), was added to the standard MRI protocol before surgery. Five patients were excluded, and in 20 patients, synthetic MR images were created from the quantitative scans. Manual regions of interest (ROIs) outlined the visibly contrast-enhancing border of the tumours and the peritumoral area. Contrast enhancement was quantified by subtraction of native images from post GD-images, creating an R1-difference-map. The quantitative R1-difference-maps showed significant contrast enhancement in the peritumoral area (0.047) compared to normal appearing white matter (0.032), p = 0.048. Relaxometry detects contrast enhancement in the peritumoral area of malignant gliomas. This could represent infiltrative tumour growth.


Materials and methods
Subjects. Twenty-five patients with radiological findings suggestive of high-grade malignant gliomas were prospectively included in the study from 2013 to 2016. This cohort has previously been reported in 25 . Patients were examined with MRI before surgery. The diagnosis of malignant glioma was confirmed with histopathological analysis after surgery. Three patients were excluded due to other diagnoses; one abscess, one lymphoma and one primitive neuroectodermal tumour (PNET). Tumour classification was made according to WHO 2007, which was clinical standard at the time of inclusions. Patient demographics are summarized in Table 1. Two patients were excluded from the analysis due to motion leading to invalid data sets. The local institutional review board of the Swedish Ethical Review Authority in Linköping approved the study (Dnr 2011/406-31), and informed written consent was obtained from all patients. MRI protocol. Magnetic resonance data was acquired on a 3 T MR scanner (MR750, GE Medical Systems, Milwaukee, Wisconsin, US) using a 32-channel phased array head coil. Patients were examined according to the clinical protocol for brain tumour investigation, with the addition of the qMRI sequence SyMRI MAGiC (GE Healthcare) both before and after contrast agent injection. The clinical protocol for brain tumour investigation consisted of conventional axial T2WI-FLAIR, T1WI, T2WI, diffusion weighted images (DWI), dynamic susceptibility contrast (DSC) perfusion, T1WI-GD, and 3D-FSPGR (fast spoiled gradient echo) GD. The sequence parameters for the conventional images used in the study analysis were as follows: T1WI spin echo before and after (T1WIGD) contrast agent injection: axial, FOV 220 × 165 mm, 24  The quantitative sequence is a multi-slice, multi-echo and multi-saturation delay qMRI technique for simultaneous measurement of R 1 , R 2 and PD 26 , with the following parameters in this study: qMRI MAGiC; axial, FOV 220 × 180 mm, 24 slices, voxel size 0.43 × 0.43 × 5 mm (gap 1 mm), ASSET acceleration factor 2. In total 8 images per slice were measured with TE = 22 ms or 95 ms, TR = 4 000 ms, TI = 170, 670, 1840 or 3840 ms. The scan time was 5:55 min, and the qMRI-series was obtained before and after contrast agent injection. The synthetic images created from the qMRI scan had the same settings as the corresponding conventional MR images. Table 1. Patient demographics and volume of manually delineated contrast-enhancing part of tumour including necrotic core in T1 3D-FSPGR GD. Patient age, sex, the histopathological diagnosis according to WHO 2007 classification for brain tumours, and the volume of contrast enhancing tumour including the necrotic core in the 3D-FSPGR GD images. www.nature.com/scientificreports/ After contrast agent injection spin echo images were acquired first followed by the 3D-FSPGR and lastly the qMRI.
Post-processing and ROI placement. qMRI post-processing. The qMRI sequence yields quantitative maps of R 1 , R 2 and PD, which are used for measurements and to create synthetic images matching the conventional images, using the same parameters as in the conventional images. The post-processing of the raw image dataset to create synthetic images was performed on a conventional computer using SyMRI 8 software (Synthet-icMR AB, Linköping, Sweden).
A neuroradiologist (IB) manually drew regions of interest (ROIs) for the analysis. The outer contrast-enhancing borders of the tumours were delineated (tumour-ROI) in the 3D-FSPGR GD images and in the synT1WI-GD ( Fig. 1A) separately (Fig. 1B). The whole tumour was delineated slice by slice.
A second free hand extended ROI (extended tumour-ROI) was placed approximately 1 cm outside of the tumour-ROIs in the synT1WI-GD. Care was taken not to include contrast-enhancing structures, e.g. vessels or the choroid plexus (Fig. 1A).
The tumour volumes were calculated from the 3D-FSPGR GD images in MATLAB (MathWorks Inc, Natick, US). The conventional images were used for volume calculation due to their higher resolution.
Using the synthetic images, the peritumoral area was analysed by subtracting the tumour ROI from the extended tumour ROI to achieve a peritumoral ROI. In some tumours with a necrotic centre, an additional ROI was placed in the necrosis (necrosis ROI).
ROIs were also placed in synT1WI and synT1WI-GD in the normal appearing white matter in the corresponding lobe at the same level in the contralateral hemisphere (NAWM ROI). The NAWM ROI-size was approximately 1 cm in diameter. Similar sized ROIs were also placed in the distal peritumoral oedema at the periphery of the oedematous zone (distal oedema ROI). Hence there are three areas that were quantitatively compared: the peritumoral ROI, the distant oedema ROI, and the NAWM ROI. The tumour ROI and the necrosis ROI are presented in the descriptive histograms.
Calculation of R 1 -difference and PD. A transformation matrix was calculated for pre-GD to post-GD transformations by registration of synT2WI to synT2WI-GD, using the MeVisLab MERIT module with registration method set to "3D rigid" and similarity measurement to "SSD" (sum of squares of intensity distance). The transformation matrix was used to transform quantitative R 1 -maps from pre-GD space to post-GD space. Then, the quantitative difference in R 1 -relaxation due to contrast enhancement was calculated as the difference between the R 1 -maps post-GD and the transformed native R 1 -maps (R 1 -diff map). The R 1 -diff map is proportional to the GD concentration in the brain according to the relationship R 1 postGD − R 1 preGD = relaxivity × GD concentration 27 .
The ROIs were then applied to the R 1 -difference map and the PD map. The mean and standard deviation were calculated for each ROI in each subject. A one-sided t-test was performed on the peritumoral ROI and the NAWM ROI; calculations were performed in Microsoft Excel.
All methods were carried out in accordance with relevant guidelines and regulations.

Results
The volumes of the contrast-enhancing portions of the tumours, including necrotic centre when applicable, were calculated from the conventional 3D-FSPGR GD and are presented in Table 1.
The tumours have a heterogeneous appearance on conventional MR-images, and also exhibit different relaxation patterns depending on the tumour structure. This is depicted in Fig. 2, which shows examples of two patients with glioblastomas with different appearances on MRI images as well as different relaxation patterns. Figure 2A,B (same patient) show a typical glioblastoma appearance with an intense contrast enhancement in the periphery, a more liquefied necrotic centre and peritumoral oedema. The contrast enhancement is depicted in the R 1 -difference graphs. For this tumour the R 1 -difference distribution has a maximum around zero corresponding to the necrotic centre (green line in 2E), and the contrast-enhancing part of the tumour contributes to the right-shifted, higher values on the x-axis (red line in 2E). Figure 2F,G (same patient) show another appearance of a glioblastoma, with a more solid tumour portion, exhibiting irregular contrast enhancement throughout the tumour and a peritumoral oedema. The R 1 -difference graph for this tumour shows tumour relaxation values in a more dispersed, right-shifted distribution (red line in 2 J). Figure 2C,H show maps of the R 1 -difference in the tumour, which corresponds to the contrast enhancing part of the tumour. Figure 2D,I show the maps of the R 1 -difference in the peritumoral ROI.
In Fig. 3 the R 1 -difference is depicted for all voxels within respective ROI. Figure 3A shows the R 1 -values before (dashed line) and after (dotted line) GD-based contrast agent in the peritumoral area, in the NAWM, and in the distal oedema. R 1 -values are lower in the peritumoral area compared to the NAWM due to the higher water content of the peritumoral oedema. The R 1 -values are the lowest in the distal oedema for the same reason. The dotted lines are more right-shifted compared to the dashed lines due to the effect of gadolinium. Figure 3B shows the contrast enhancement as the R 1 -difference for all tumours; in the peritumoral area (blue), in the NAWM (orange), and in the distal oedema (purple). The R 1 -difference overlaps, but the mean R 1 -difference is significantly higher in the peritumoral oedema compared to the NAWM (p = 0.048).
All R 1 -difference curves have values below zero, which would not be expected after injection of gadolinium based contrast agent. This is probably due to the distribution of the inherent noise in the data, but it could also partly be due to movement of the patient. However, movement of the patient would lead to both positive and negative R 1 -difference values and in mean the effect this will be cancelled out. Maps of the R1-differences, positive and negative, for the tumors and the peritumoral areas are provided in the supplementary figures. In Fig. 5, paired maps of the R 1 -difference of the tumour (left) and the peritumoral ROI (right) are shown for all patients. The threshold is set at values higher than group mean NAWM + one standard deviation.
Relaxation values before and after gadolinium based contrast agent injection in the peritumoral area, the distal oedema and the NAWM are shown in Table 2. There is a significant difference in R 1 in all three tissues, indicating an effect of the contrast agent outside of the contrast enhancing part of the tumour. Table 3 shows the mean and standard deviation (SD) of PD and of the R 1 -difference (R 1 before gadolinium subtracted by R 1 after gadolinium) in the ROIs of the peritumoral-ROIs compared to the NAWM-ROIs in synthetic images. There is a significant difference in the R1-difference between the peritumoral area compared to the NAWM (p = 0.048), with a higher R 1 -difference in the peritumoral area indicating a greater T1-shortening effect after gadolinium based contrast agent injection here, compared to NAWM. There is also a significant difference in PD value between the peritumoral area and the NAWM (p < 0.001), with a higher PD value in the peritumoral area as expected, due to higher water content in the oedema than in the NAWM.

Discussion
Visual assessment of the peritumoral area of malignant gliomas is a radiological challenge, since the diffuse, non-enhancing infiltration of these brain tumours cannot be visualized using conventional MRI 28 . The nonenhancing parts of the tumours are also a challenge during follow-up, since they have to be considered in the treatment evaluation 29 but are difficult to detect. Clinically, this transitional zone between the contrast-enhancing part of the tumour and the peritumoral oedema is therefore of great interest, since the surgical treatment aims to resect as much as possible of the tumour without causing impairing sequelae to the patient. However, even with radical resection, tumour recurrence is in most cases localized to the resection margin 30 , and at present there is no conventional MRI technique implemented in standard practice to visualize the non-enhancing parts of the malignant gliomas. The MRI-technique used in this study is quantitative compared to conventional MRI,  www.nature.com/scientificreports/ which relies on visual assessment. The benefits of a quantitative technique is that it is measurable and provides a basis for longitudinal comparison, since it is independent of scanner settings, inhomogeneity of the B1-field, and coil sensitivity profile.
In this study we report that it was possible to measure contrast enhancement in the peritumoral area, evident as a significant R 1 -difference after gadolinium, which could indicate that there is a subtle blood-brain barrier leakage present. This observation may be a result of diffuse tumour infiltration with a subsequently increased neo-angiogenesis, a process that also affects perfusion properties of the peritumoral tissue 31 .
These findings are in line with other studies. Quantitative MR sequences, such as T1-and T2-relaxometry, have been shown to have the capacity to detect non-visible tissue changes in the peritumoral oedema in malignant gliomas prior to surgery 24 . Moreover, such techniques have also been used to detect early tissue changes in glioblastoma recurrence during follow-up, before any changes were evident in the conventional images 32,33 . Additionally, the finding of non-visible peritumoral contrast enhancement is in line with the findings of Müller et al. 34 , using quantitative T1-mapping to detect peritumoral contrast enhancement invisible in standard MRI. They also showed that a decrease of this invisible contrast enhancement during treatment was prognostic for a favourable treatment response.
Ellingson et al. used conventional T1W GD subtraction maps with an automated method for detection of contrast enhancement. They examined patients during follow-up of glioblastomas and found an improvement in delineation and prediction of survival compared to conventional segmentation 35 . Conventional images, however, have the drawback of arbitrary intensity scaling, which makes it difficult to compare results between different scanners and different patients, a problem that can be solved by using a quantitative sequence, which actually measures the relaxation values of the tissue, like in this study.
Even in regions with an intact blood-brain-barrier there is some T1-shortening in the NAWM of the brain parenchyma following GD based contrast agent injection 36,37 . This physiological phenomenon is evident also in the values of NAWM and in the distal oedema in this study, which show an R 1 -difference after GD based contrast agent injection, but to a lesser extent than in the peritumoral area.
This study has some potential limitations. We hypothesize that our findings of an increase in peritumoral R 1 -difference could be due to infiltration of tumour into the peritumoral oedema. However, this study lacks the histopathological proof of correspondence with tumour infiltration.
When subtraction maps are used, the performance of the image registration procedure is important. The registration of quantitative R 1 maps therefore benefits from the registration to synthetic T2WI, which to a large degree are not very affected by GD. Thus the registration of the quantitative values is, in our view, more robust compared to registration of the corresponding conventional T1WI. A limitation in this study is therefore that any subject motion will reduce the effective spatial resolution due to the influence of partial volume effect in the registration, especially in the z-direction (i.e. head-feet) because of the relatively thick slices that were used. Recently, a method for synthesizing a contrast-enhancement map based on post-contrast images only has been suggested 38 . This novel procedure would overcome the problem of possible patient motion and therefore make the technique even more robust.
Another challenge is that the tissue changes that we are interested in detecting are minimal and inhomogeneous, and it is therefore a major challenge to distinguish the pathology from the normal tissues, which also creates noise/variance during MR examination. This is perhaps most evident from the R 1 -difference graphs, in which the ROIs from necrosis and NAWM have slightly negative average difference-values, which is not intuitive, but is due to the inherent noise level.
The small number of patients is a limitation of this study, and the tumour classification is not made according to the latest WHO standard since it was not implemented at the time of patient recruitment. Since the inclusion criteria were based on visual assessment of a contrast-enhancing tumour with malignant appearance, the patient Table 2. R 1 before and after GD in peritumoral area, distal oedema, and NAWM. R 1 values before and after GD-based contrast agent injection in the peritumoral area, the distal oedema, and in the NAWM.

Pre-GD
Post-GD  www.nature.com/scientificreports/ group is heterogeneous, ranging from the grade IV WHO tumours, to a grade II tumour. For future studies, it could be of interest to investigate how relaxation properties are correlated to new molecular markers, e.g. IDHstatus. The analysis could also benefit from a quantitative sequence with isotropic voxels and thinner slices, for improved segmentation. This could aid in the continuing the work, to visualize these changes on an individual level.
In conclusion, the use of quantitative subtraction maps based on relaxometry makes it possible to detect contrast enhancement in the peritumoral oedema of malignant gliomas. We interpret this finding as a quantitative representation of tumour infiltration and the technique described here could therefore potentially be useful in the context of surgical planning. The procedure could then aid maximum resection of the tumour, while preserving areas affected only by peritumoral oedema, thereby potentially improving the outcome for the patient.

Data availability
Datasets analysed in this study are not available due to patient confidentiality.