Fast and Quantitative T1ρ-weighted Dynamic Glucose Enhanced MRI

Common medical imaging techniques usually employ contrast agents that are chemically labeled, e.g. with radioisotopes in the case of PET, iodine in the case of CT or paramagnetic metals in the case of MRI to visualize the heterogeneity of the tumor microenvironment. Recently, it was shown that natural unlabeled D-glucose can be used as a nontoxic biodegradable contrast agent in Chemical Exchange sensitive Spin-Lock (CESL) magnetic resonance imaging (MRI) to detect the glucose uptake and potentially the metabolism of tumors. As an important step to fulfill the clinical needs for practicability, reproducibility and imaging speed we present here a robust and quantitative T1ρ-weighted technique for dynamic glucose enhanced MRI (DGE-MRI) with a temporal resolution of less than 7 seconds. Applied to a brain tumor patient, the new technique provided a distinct DGE contrast between tumor and healthy brain tissue and showed the detailed dynamics of the glucose enhancement after intravenous injection. Development of this fast and quantitative DGE-MRI technique allows for a more detailed analysis of DGE correlations in the future and potentially enables non-invasive diagnosis, staging and monitoring of tumor response to therapy.

As the origin of the DGE contrast is still under discussion [1][2][3][4][5][6][7][8][9][10][11] , the presented technique does not only form a simple and robust diagnostic tool for studying the DGE contrast in clinical studies, but with its high temporal resolution also serves as a research tool. Thus, it might help solving the question to what extent the occurring contrast originates from intra-or extracellular glucose level changes, from pH changes, or from other glucose related metabolites.

Results
T 1ρ -weighted DGE-MRI in vivo. The accelerated and quantitative T 1ρ -w DGE-MRI protocol optimized with respect to contrast and scanning time was applied with temporal resolution of less than 7 seconds in the study of a patient with a brain tumor. The tumor (glioblastoma, WHO grade IV) located in the left frontal lobe can be identified in the T 2 -w image acquired at 7T (Fig. 1a) and in the co-registered gadolinium contrast-enhanced T 1 -weighted (GdCE-T1w) image obtained at 3T (Fig. 1b). We define the quantitative T 1ρ -weighted dynamic glucose enhancement (DGE ρ ) by the relative signal difference (Eq. 4 in Methods) at each time point  ). First of all, the DGE ρ images obtained after glucose injection (cf. Fig. 1c, t = 588 s) clearly delineate the tumor region consistent with the GdCE-T1w image (Fig. 1b). We further evaluated DGE ρ as a function of time in three regions of interest (ROIs), namely a tumor-ROI (ROI #1) selected on the DGE ρ image shown in Fig. 1c, a second tumor-ROI (ROI #2) selected on the GdCE-T1w image (Fig. 1b), and a ROI in normal appearing white matter (ROI #3). The ROI-specific DGE ρ curves are shown in Fig. 1d. The ROIs are marked in the GdCE-T1w and DGE ρ image shown in the top left corner. Before the start of glucose injection (at t = 0 s) DGE ρ of all three ROIs fluctuated around 0% and consequently no tumor contrast was visible in the corresponding DGE ρ image shown in Fig. 1e. After the start of injection, all curves slightly increased and a faint contrast became apparent in the tumor area as well as in the paraventricular area about 1 min after the end of the injection phase (Fig. 1f). For t ≥ 200 s the curves of both tumor ROIs were outside the error of normal appearing white matter (ROI #3), which showed only a minor increase in the DGE ρ curve over the entire time course. Accordingly, the contrast in the DGE ρ images increased, revealing another slightly enhancing region (black arrow) at the bottom of the tumor area (Fig. 1g), which remained visible in the DGE ρ images obtained afterwards ( Fig. 1h and i). The highest contrast was observed at about 10 min after start of the injection (Fig. 1h), where the T 1ρ -w dynamic glucose contrast in ROI #1 was more than twice as high compared with that in ROI #2 and about 8 times higher than in normal appearing white matter. The subsequent signal drop in the DGE ρ curve (red arrow in Fig. 1d) was most likely due to patient motion, which was identified by a displacement of the brain position in the time-resolved T 1ρ -w images. Interestingly, the DGE ρ images did not show any contrast in blood vessels.

Bloch-McConnell simulations.
To investigate the contrast obtained with T 1ρ -w MRI we simulated T 1ρ relaxation curves by means of a Bloch-McConnell simulation tool. Figure   This yields the relative signal change Δ S rel given by equation (4), that also defines DGE ρ (t) (Eq. 1) at a specific time point t. Δ S rel , and thus DGE ρ (t) depends only on TSL and the variation Δ R ex of the exchange-dependent relaxation rate R ex . In Fig. 2c Δ S rel is plotted as a function of the glucose concentration change Δ c Glc for R 2 = 15 s −1 (solid blue line) and R 2 = 25 s −1 (green diamonds) for one specific spin-lock time of 50 milliseconds. The plot shows that the contrast does not depend on R 2 . The analytical approximation (Eq. 4; dashed red line) agrees well again. Thus, DGE ρ (t) defined by Δ S rel represents a quantitative contrast, which depends linearly on Δ R ex and hence on changes of the glucose concentration (Δ c Glc ) for a given TSL. To determine the optimum TSL (TSL opt ) one has to consider not only the signal-to-noise ratio (SNR), but also the contrast-to-noise ratio (CNR). Assuming a constant SNR, the maximum CNR is given by the position of the maxima of Δ S. Equation (3) allows to determine this point analytically yielding TSL opt = T 1ρ . For the simulated relaxation rates, which represent the range we observed in human brain tissue at 7T using the adiabatically prepared spin-lock approach 11 , the CNR for TSL = 50 ms (red vertical line in Fig. 2b) is close to the optimal value for all relaxation rates that we considered.

In vitro experiments.
To confirm the results of our simulations we performed measurements of aqueous solutions with different glucose concentrations and different R 1 and R 2 . The relaxation rates were adapted using gadoteric acid and agar for one set of solutions and MnCl 2 for a second set. In the following, the different sets are called agar phantoms and MnCl 2 phantoms, respectively. The measured T 1ρ relaxation curves for glucose concentrations of 20 mM (solid lines) and 40 mM (dashed lines) are plotted in Fig. 3a for the agar and the MnCl 2 phantoms. The curves were normalized to the first value and all data points represent the mean and standard deviation of three independent measurements. Figure 3b shows the signal differences Δ S between the two particular relaxation curves from Fig. 3a. The curves display the dependence on the absolute relaxation rates expected from the simulations (Fig. 2b) and equation (3). Figure 3c shows the relative signal difference Δ S rel for a constant spin-lock time of 100 milliseconds as a function of the glucose concentration change. The curves for both, the agar and the MnCl 2 phantoms, agree within the errors (shown only for the MnCl 2 phantom measurements for the sake of clarity). The consistency of both curves proves the independence of the relative signal difference Δ S rel (or rather DGE ρ for series of in vivo images) on absolute relaxations rates.

Discussion
In this study, we showed that T 1p -based DGE-MRI can be accelerated essentially by employing T 1ρ -w imaging. The introduced contrast called T 1ρ -w dynamic glucose enhancement (DGE ρ , Eq. 1) was shown to be independent of relaxation parameters of tissue and direct proportional to changes of the glucose concentration thus enabling fast and quantitative DGE-MRI in a glioblastoma patient with a temporal resolution of less than 7 seconds.
So far glucose enhanced MRI in humans has been performed in brain tumor patients at 7T by means of CEST 9 and T 1ρ mapping 11 and in head and neck tumor patients at 3T by means of CEST 10 . In all studies, an increased glucose uptake was reported after intravenous injection of natural D-glucose. However, the studies substantially differed in the temporal resolution, varying between 5 seconds in the case of CEST-based dynamic glucose enhanced MRI applied by Xu et al. 9 and about 5 minutes in the study of Wang et al. 10 . The temporal resolution of the T 1p -weighted approach proposed in this study is below seven seconds and thus in the same order as for CEST-based DGE-MRI. As the spin-lock preparation time of 50 ms is much shorter compared to CEST saturation, which normally requires seconds, the temporal resolution can be increased to about 3 s if SNR is sufficient.
High temporal resolution is mandatory to detect variations on small time scales like changes in the blood glucose level (BGL) after a bolus glucose injection. Robust tracking of the BGL could potentially enable pharmacokinetic modelling based on compartment models as for example employed in gadolinium-based dynamic contrast enhanced MRI (DCE-MRI) 12 . Another benefit of a high temporal resolution is the opportunity to increase the effective SNR and CNR by averaging of several consecutive measurements. This could be relevant for glucose enhanced MRI when a lower temporal resolution is sufficient, e.g. when the bolus injection is replaced by a continuous glucose infusion, but also for native T 1ρ -based imaging without glucose enhancement. Consequently the presented adiabatically prepared T 1ρ -w imaging technique with the proposed normalization might also improve cartilage imaging, where T 1ρ mapping is a common technique to detect the loss of proteoglycan in the early stages of osteoarthritis [13][14][15][16] .
As shown previously, an adiabatically prepared spin-lock approach combined with a non-adiabatic MRI readout, as used in our study, works within specific absorption rate (SAR) restrictions and technical limitations of ultrahigh field whole-body scanners 11 . This leads to a homogenous T 1ρ contrast over the entire brain despite B 1 inhomogeneities and consequently to negligible contributions from B 1 dispersion to the DGE contrast 11 . We want to point out that for the in vivo T 1ρ -w DGE-MRI measurement, SAR was around 50% of the allowed value and hence relatively low for using adiabatic pulses. This can be understood since only two adiabatic half-passage pulses are used per 7 s. Consequently, a reduction of the recovery time and thus an increase of the temporal resolution is also in accordance with SAR restrictions. Furthermore, the proposed T 1ρ -w DGE-MRI inherits all benefits of the adiabatically prepared spin-lock approach. This includes the higher sensitivity to the intermediate and fast exchange regime relevant for glucose and the enhanced robustness against B 0 inhomogeneities compared to CEST 6,17 , but also the fact that changes in DGE ρ due to inhomogeneities in the B 1 field are negligible compared to changes induced by variations of the glucose concentration 11 . Especially the robustness against field inhomogeneities qualifies the presented approach for application at whole-body ultra-high field scanners. These are of great interest for chemical exchange sensitive experiments due to the increasing exchange-weighting with higher field strength 18,19 . Further, the robustness against field inhomogeneities makes the application of correction methods dispensable and thus simplifies the post-processing.
As predicted by our analytical approximation (Eq. 3) we could show that the dependency of the signal difference (Δ S) on absolute relaxation rates can be eliminated by an appropriate normalization yielding the T 1ρ -weighted dynamic glucose enhancement (DGE ρ ), which depends linearly on the glucose concentration and is independent of relaxation parameters of the tissue. These properties could be verified with simulations (Fig. 2c) and in vitro measurements (Fig. 3c). We want to point out that the intrinsic robustness of the adiabatic spin-lock against field inhomogeneities in combination with the introduced normalization yield a quantitative contrast, which can be compared between different measurements and subjects. CEST-based DGE-MRI techniques, on the other hand, can be prone to influences of inhomogeneities and absolute relaxation rates 20 . Although faster T 2 relaxation due to exchange 21 can lead to a signal enhancement in CEST-based DGE-MRI, being independent of T 1 and T 2 relaxation has the benefit of the above mentioned quantitative contrast and additionally some practical benefits: with DGE ρ it is possible to perform a DGE measurement after gadolinium injection which is practical in clinical routine. Beyond that, it is also thinkable to perform DGE and DCE with the same injection bolus at the same time, which would speed up the acquisition and provides a reference for pharmacokinetic investigations.
However, also in the case of CEST the influences of absolute relaxation rates can be handled by employing relaxation compensation techniques. Figure 4 shows simulated T 2 -and T 1 -dependencies of the CEST-based DGE approach based on the paper of Xu et al. 5 . Our simulations reveal that by using a T 1 map and the AREX 22 metric, also quantitative CEST-based DGE-MRI can be realized. DGE ρ in brain tumor patient. Applying the proposed quantitative T 1ρ -w DGE-MRI approach with seven-second temporal resolution evaluated using DGE ρ in a glioblastoma patient we observed an increasing DGE contrast in the tumor area after the intravenous glucose bolus injection. This finding is in agreement with the outcome of our previous DGE-MRI study of a glioma patient based on T 1ρ mapping 11 and the in vivo study of Xu et al. 9 employing CEST-based DGE-MRI in brain tumor patients. A quantitative evaluation of DGE ρ in three regions of interest (ROIs) revealed a substantially increased contrast in the tumor ROIs selected on the DGE ρ and GdCE-T1w images compared to normal appearing white matter. Interestingly, the hyperintense tumor areas in the DGE ρ images (cf. Fig. 1f-i) partially overlap but still differ from those on the GdCE-T1w image (Fig. 1b). The observed difference in both contrasts is in agreement with the findings of Walker-Samuel et al. 3 , who did not observe a significant correlation between glucoCEST and GdCE-T1w contrast in an animal study. This allows for the conclusion that DGE-MRI can provide complementary information about pathologies compared to contrast enhanced T1-w MRI, which is the current gold standard method for detecting and characterizing high-grade glioma 23 by visualizing blood brain barrier (BBB) disruption. We could not validate whether the enhancing region outside the tumor area (black arrow; Fig. 1g), which was not visible in the native T 2 -w and GdCE-T1w images ( Fig. 1a and b) was an active tumor region or not. Hence, it remains to be shown if DGE-MRI might highlight hidden active regions of the tumor and thus forms a tool for the early detection of cancer. Whereas CEST-based DGE-MRI showed an uptake in blood vessels 9 , this was not observed by T 1p -w DGE-MRI. It remains to be investigated in detail if this is due to the short saturation period of spin-lock compared to CEST or if it has a meaning on the contrast origin level.
Origin of DGE ρ contrast. We showed that the DGE p signal increases in tumors. However, the actual origin of the signal changes in DGE-MRI is still under discussion 1-11 . Chan et al. 1 stated that the signal in glu-coCEST originates mostly from the extracellular compartment, and, due to lower pH, predominantly from the extracellular-extravascular glucose. Further, Chan et al. 1 as well as Walker-Samuel et al. 3 showed that FDG-PET and glucoCEST MRI are enhancing similarly. In contrast to Chan et al., Walker-Samuel et al. concluded from the similarity with FDG-PET that also intracellular compounds contribute to the glucoCEST signal. This conclusion was also based on their results of 13 C spectroscopy after injection of 13 C labeled glucose that showed appearance of glucose, glucose-6-phsophate, fructose phosphates, as well as amino acids such as glutamate, glutamine, taurine and alanine. From phantom experiments they further conclude that glucose and its metabolic products as well However, employing R 1 = 1/T 1 and the apparent exchange-dependent relaxation evaluation AREX = (S 0 /S ref − S 0 /S) · R 1 also CESTbased DGE-MRI yields a relaxation independent contrast. The simulated CEST pre-saturation parameters were chosen similar to Xu et al. 9 : 32 sinc-gauss pulses (50 ms, Δ ω = 1.2 ppm, B 1 = 1.96 μ T, separated by a 25 ms delay, each) and a delay of 2 s after each scan. The water and solute pool parameters were chosen similar to the CESL simulations in Fig. 2. as glutamate and glutamine might contribute to the glucoCEST signal, but lactate protons are exchanging too fast to be detectable with CEST 3 . For the case of glucoCESL, Jin et al. 6 also mention the contribution of glucose metabolism products. Thus, it is still under discussion to what extent DGE-MRI is extracellular and consequently only with indirect access to metabolism, or intracellular, which would give more insight to metabolism. From our data, we can only conclude that changes in T 1ρ -based DGE-MRI originate from a different compartment than in gadolinium enhanced MRI, which is coherent with both origins, the extracellular extravascular and the intracellular space or a mixture of both. This conclusion is also in coherence with previous publications 1, 3,5,11 . However, in accordance with Jin et al. 6 we want to point out that with on-resonant T 1ρ -based DGE-MRI all exchanging sites contribute to the signal and, compared to CEST, also the close to water resonating and faster exchanging pools such as lactate have a stronger contribution, as sensitivity of spin-lock to high exchange rates is improved 6,17 . As the presented technique can track the signal changes fast and accurate, it might become an important tool for further investigations of the origin of the DGE contrast.
Having shown that our contrast is quantitative, we can employ the in vitro calibration to try calculating the corresponding glucose concentration in vivo similar to Jin et al. 6 . Assuming the relaxivity measured in phantoms (Fig. 3) to be valid also in vivo, the obtained DGE ρ or rather change of R 1ρ in the tumor would correspond to a glucose concentration increase of up to 40 mM (721 mg/dL), using the relaxivity reported by Jin et al. 6 the concentration change would be approximately 25 mM (450 mg/dL). Although Xu et al. measured a venous glucose level of up to 23.7 mM (427 mg/dL) in volunteers about 2-4 min after the injection 9 , a value between 25 mM and 40 mM still seems to be improbably high. This hints that the observed signal change might not solely originate from the hydroxyl exchange of glucose, but as discussed above, also from glucose metabolic products 3,6 , and glutamate and glutamine 3 . Moreover, the relaxivity of the DGE effect potentially differs between the in vivo and in vitro situation as it depends on temperature, pH, and the concentration of exchange catalysts and has not yet been determined directly in vivo or even in tumors.

Unexpected signals and motion correction.
After injection, glucose is also expected to rapidly enter the cerebrospinal fluid (CSF) leading to an increase of R 1ρ and consequently to a positive DGE contrast in the ventricles as observed in our measurements. However, it has also been reported that a glucose injection results it volumetric changes of the CSF compartments 24 . This is an explanation for the negative contrast observed in the ventricles by Xu et al. employing CEST-based DGE-MRI 9 . Such volumetric changes can also lead to a reduction of R 1ρ and consequently to a negative DGE contrast employing T 1ρ -weighted DGE-MRI explaining the observed signals in the outer CSF compartments, where pixels are expected to be affected by partial volume effects, which most likely result from the limited special resolution in z-direction. Volumetric changes of the CSF lead to an increase of the CSF fraction in the partial volume affected voxels and consequently to negative DGE contrasts, as R 1ρ of CSF is about one order of magnitude smaller than for brain tissue 8,11 . This insight must be included when interpreting DGE uptake of tumors close to CSF regions.
Generally, patient motion is a problem of every contrast based on signal differences between different time points, including all CEST-and CESL-based DGE-MRI approaches, but also dynamic contrast enhanced (DCE) MRI or functional MRI (fMRI). For correction of motion after data acquisition, we employed a rigid registration algorithm. However, for a robust post-process correction of extensive out-of-plane motion the acquisition of an expanded volume is mandatory; for example by applying single-shot 3D MRI sequences such as 3D gradient echo-based MRI 25 or echo-planar imaging (EPI) speed-up 26 , which can easily be combined with the T 1ρ -weighted preparation 11 . An alternative method to reduce patient motion is the application of immobilization devices known from radiation therapy as done by Wang et al. 10 . In principle, also a combination of post-process motion correction and immobilization of the patient is possible.
In conclusion, dynamic glucose enhanced MRI (DGE-MRI) might open the window to non-invasive observation of glucose uptake and potentially metabolism. Due to its high temporal resolution in combination with a high robustness against field inhomogeneities and a high sensitivity to glucose, T 1ρ -weighted DGE-MRI has a high potential to facilitate the translation of glucose enhanced MRI into the clinics. The simple quantitative evaluation can be performed online directly at the scanner to fulfill the clinical demand for practicability. Quantitative DGE further allows a deeper insight into the underlying correlations and in principle enables combined measurements with relaxation affecting contrast agents such as Gd. Further longitudinal studies with larger numbers of patients with different tumor grades are planned to investigate the full potential for detection and staging of cancer or also neurodegenerative diseases by means of the proposed fast and quantitative T 1ρ -w DGE-MRI technique.

Methods
R 1ρ theory and glucose contrast. For a two-pool system (one water, one solute proton pool) the on-resonant longitudinal relaxation rate in the rotating frame R 1ρ is given by 27 R 1ρ = R 2 + R ex where R 2 is the transverse relaxation rate of water protons without contributions from chemical exchange and R ex the exchange dependent relaxation rate. R ex can be approximated as 6 where Δ p is the ratio of concentrations of solute and water protons, k B is the exchange rate (units of s −1 ) and δ the resonance shift (units of rad/s) between the solute and water proton pools, and ω 1 = γ B 1 is the amplitude of the spin-lock pulse (units of rad/s). For T 1ρ -weighted MRI we could show that the difference in signal intensities (ΔS) between a voxel and a reference voxel with different exchange-dependent relaxation, e.g. gray and white brain matter, can be approximated as 11 Scientific RepoRts | 7:42093 | DOI: 10.1038/srep42093   Patient examination. As part of a clinical study T 1ρ -w DGE-MRI was applied in the examination of a 66-year-old male patient with newly diagnosed and histopathologically confirmed glioblastoma (WHO grade IV). The examination was approved by the local ethics committee of the Medical Faculty of the University of Heidelberg and is in accordance with the relevant guidelines and regulations. Written informed consent was received from the patient prior to the examination. The patient was examined after a 6-hour fasting period ensuring a normal blood glucose level before injection. Using an intravenous line 100 ml of 20% D-glucose (SERAG-WIESSNER GmbH & Co. KG, Naila, Germany) were injected manually over 2 min into an arm vein. Two blood samples were taken, one before and the other approximately 25 min after the glucose injection. The blood sugar values, determined by means of a conventional blood sugar meter (Accu-Chek Aviva; Roche Diagnostics, Rotkreuz, Switzerland), were 106 mg/dL (5.9 mM) and 146 mg/dL (8.1 mM) pre-and post-injection, respectively. The complete protocol of the patient examination is sketched in Fig. 5a. The total examination time, including patient preparation and positioning, morphological and T 1ρ -w DGE MRI, as well as the blood glucose measurements, was approximately 60 min. The T 1ρ -w DGE-MRI part shown in Fig. 5b consisted of n = 178 individual measurements leading to an acquisition time of about 20 minutes. The first 18 measurements were performed before the start of the glucose injection and yielded the reference for the calculation of the dynamic glucose enhancement (Eq. 1).
Data acquisition and analysis. All MR measurements were performed on a 7T whole-body MR scanner (MAGNETOM 7T, Siemens Healthcare, Erlangen, Germany) using a 24-channel head coil (Nova Medical, Wilmington, MA, USA). The MR sequence used for T 1ρ -based MRI consists of an adiabatically prepared spin-lock pulse cluster as described in Schuenke et al. 11 and shown in Fig. 5c followed by a conventional MRI readout. The parameters of the adiabatic hypsec-pulses were: RF amplitude B 1, max ≥ 20 μ T, adiabatic sweep time t adia = 8 ms, bandwidth Δ = 1200 Hz, and μ = 6, where μ is a dimensionless parameter that controls the pulse shape 29 . The spin-lock frequency was adjusted manually to obtain the desired value of B 1 ≈ 5 μ T in the region of interest, e.g. the tumor area. For MRI readout we used a centric-reordered single-shot gradient echo (GRE) sequence. In vivo we acquired three axial slices in an interleaved way (matrix = 128 × 104, FoV = 220 × 178 mm 2 , T E = 3.61 ms, T R = 23 ms, flip angle = 10°, slice thickness = 5 mm, distance factor = 20%). The same MR sequence with an adapted preparation block was used to obtain B 1 maps by means of the WASABI 30 approach.
In the patient examination we further acquired a stack of 32 high-resolution (0.4 × 0.4 × 2 mm 3 ) T 2 -weighted images using a Turbo-Spin-Echo (TSE) sequence (T E = 52 ms, T R = 12340 ms). The Gadolinium contrast-enhanced T 1 -weighted (GdCE-T1w) images were acquired 10 days prior to the 7T examination in the course of a clinical MR protocol at 3T. The GdCE-T1w, T 2 -w and T 1ρ -w images were co-registered and the slice thickness of the GdCE-T1w and T 2 -w images was interpolated to the slice thickness (5 mm) of the T 1ρ -w images using a multi modal rigid registration algorithm in the DKFZ Image Processing Platform -an in-house version of the Medical Imaging Interaction Toolkit (MITK) 31 . All further post-processing and data analysis, including a rigid in-plane motion correction of the T 1ρ -w images was performed using self-written software in MATLAB. All errors were calculated taking into account the law of error propagation.