Towards robust glucoCEST imaging in humans at 3 T: Arterial input function measurements and effects of infusion time

Abstract

Dynamic glucose-enhanced (DGE) magnetic resonance imaging (MRI) has shown potential for tumor imaging using D-glucose as a biodegradable contrast agent. The DGE signal change is small at 3 T (around 1%) and accurate detection is hampered by motion. The intravenous D-glucose injection is associated with transient side effects that can indirectly generate subject movements. In this study, the aim was to study DGE arterial input functions (AIFs) in healthy volunteers at 3 T for different scanning protocols, as a step towards making the D-glucose weighted chemical exchange saturation transfer (glucoCEST) protocol more robust. Two different infusion durations (1.5 and 4.0 minutes) and saturation frequency offsets (1.2 and 2.0 ppm) were used. The effect of subject motion on the DGE signal was studied by using motion estimates retrieved from standard retrospective motion correction to create pseudo-DGE maps, where the apparent DGE signal changes were entirely caused by motion. Furthermore, the DGE AIFs were compared to venous blood glucose levels. A significant difference (p=0.03) between arterial baseline and post-infusion DGE signal was found after D-glucose infusion. The results indicate that the measured DGE AIF signal change depends on both motion and blood glucose concentration change, emphasizing the need for sufficient motion correction in glucoCEST imaging. Finally, we conclude that a longer infusion duration (e.g. 3-4 minutes) should preferably be used in glucoCEST experiments, since it can minimize the glucose infusion side effects without negatively affecting the DGE signal change.

Introduction

Chemical exchange saturation transfer (CEST) MRI and chemical exchange sensitive spin lock (CESL) MRI, using D-glucose^1-8 or glucose analogs^9,10 as a contrast agent have shown potential in tumor imaging. The CEST principle allows for a compound present in low concentrations to be detected through its chemical exchange with water after radio-frequency saturation of the exchanging protons^11,12. In a typical glucose CEST (glucoCEST) experiment, glucose is administered intravenously, followed by registration of the change in MR signal over time, an approach referred to as dynamic glucose-enhanced (DGE) MRI^2,3,7,8.

GlucoCEST has been proposed as a future alternative or complement to perfusion imaging using gadolinium since information about microvasculature, blood-brain barrier permeability and D-glucose uptake can be obtained^2,3,8,13,14. Knutsson et al.¹³ showed that it is feasible to measure arterial input functions (AIFs) using glucoCEST data acquired in healthy humans at 7 T. The CEST effect size benefits from higher magnetic field strengths, such as 7 T, since T₁ of tissue and blood water protons increases with magnetic field strength^15,16, allowing the saturation to remain for a longer period. Additionally, the chemical shift in hertz increases with field strength, which enhances the separation of the D-glucose hydroxyl proton resonances and the water protons, thus reducing the interference from direct water saturation. At lower field strengths, the fast exchange between the hydroxyl protons in D-glucose^17,18 and water protons leads to coalescence of the hydroxyl and water proton resonances, further reducing the signal change^18-20. Hence, while translation to lower clinical field strengths is ongoing and relevant, it is challenging due to the lower CEST effect.

The exchanging hydroxyl protons in D-glucose attenuate the water signal when saturation is applied to any of their resonance frequencies (from approximately 1 to 3 ppm relative to water). Subtraction of images acquired after D-glucose administration from the baseline images, enables detection of the signal difference caused by the increase in D-glucose concentration. This approach will, in contrast to magnetization transfer ratio (MTR) asymmetry analysis, be enhanced by the D-glucose concentration-dependent increase in transverse relaxation rate^21,22, which is a favorable aspect of DGE imaging. Another advantage with this difference method is that only saturation at a single offset frequency is required, which will increase temporal resolution compared to dynamic collection of a full or even partial Z-spectrum.

According to simulations, the arterial DGE signal change is expected to be around 1% at 3 T^18,20, depending on saturation strength subject to B₁ and frequency offset. Patient movement can induce signal changes of the same order of magnitude as the expected CEST signal at tissue boundaries leading to a “pseudo-CEST effect”²³. Motion correction and dynamic B₀ correction are therefore important issues to be resolved for reliable DGE imaging^20,23-26.

D-glucose is promising as a biodegradable contrast agent, but an intravenous D-glucose injection does not come entirely without (transient) side effects. Both sensory effects and physiological responses have been reported^13,20. The sensory effects reported in previous glucoCEST studies include sugary taste in the mouth, a warm or pulsating feeling at the injection site and in the head and crotch, as well as an urge to urinate^13,20. Physiological responses include vessel dilatation and volumetric changes of the ventricles²⁷. The sensory side effects are unpleasant for the subject and may, as a consequence, generate movements. The physiological changes can produce DGE signal changes that are not related to the CEST effect, which complicates the DGE MRI interpretation.

The typical D-glucose infusions used in previous glucoCEST/CESL experiments have ranged from around 30 seconds to 2 minutes in duration^2,7,8,13, using doses of either 50 mL of 50% w/v dextrose^2,13 or 100 mL of 20% w/v dextrose^7,8. In a recent glucoCEST experiment, Kim et al.²⁶ used a clamp infusion of 20% dextrose based on the DeFronzo method²⁸. By adjusting individual D-glucose infusion rates, the blood glucose level was raised to hyperglycemic levels and stabilized. This infusion method provides high stability of blood glucose levels but is time consuming. The time required to reach the targeted plasma glucose level was up to 12 minutes²⁶. An earlier study on insulin response²⁹ showed that a slower D-glucose infusion rate gives a reduced insulin response without affecting the maximum blood glucose level. While a fast injection of contrast agent is often desired for kinetics assessment in bolus tracking studies³⁰, for example dynamic susceptibility contrast (DSC) MRI, it is of interest to investigate whether a 4-minute infusion duration could reduce unwanted injection-related effects, without deteriorating the DGE image quality, as a step towards an optimized glucoCEST protocol.

The aim of this study was to collect glucoCEST images using different scanning protocols at 3 T, including two different infusion durations (1.5 and 4.0 minutes) and saturation frequency offsets (1.2 and 2.0 ppm) and to compare the arterial DGE signal change to the change in venous blood glucose level in healthy volunteers.

Methods

Subjects and MR image acquisition

All data acquisitions were performed at Skåne University Hospital, Lund, Sweden. The project was approved by the local ethics committee (The Regional Ethical Review Board in Lund, EPN 2017/673) and written informed consent was obtained from all subjects. Exclusion criteria were sensitivity for D-glucose (diabetes mellitus, sickle cell disease and blood iron deficiency). Nine healthy volunteers (2 male, 7 female, age 22-51 years) were scanned on a 3 T MAGNETOM Prisma (Siemens Healthcare, Erlangen, Germany) with a 20-channel head coil (Siemens). Morphologic images of each healthy volunteer (subject) were examined by an experienced neuroradiologist (PCS) to exclude pathology.

Seven subjects fasted for 4-6 hours before the examination as recommended to stabilize the baseline blood glucose and insulin levels. Peripheral venous catheters (PVCs) were inserted in both arms and plastic tubes were connected to the PVCs. A baseline blood glucose level was measured at arrival to assure normal fasting blood glucose levels between 3.9 and 7.5 mM (70-135 mg/dL). An intravenous infusion of 50 mL of D-glucose (50% dextrose) was performed in one arm over a period of 4.0 minutes using a power injector. The D-glucose infusion was followed by a saline flush of 30 ml, included in the total infusion duration to assure that all D-glucose entered the vessel at the same rate. Venous blood samples (~1–2 ml per sample) were collected in the contralateral arm before and 1.0, 1.5, 2.0, 3.0, 5.0, 7.0, 10, and 15 minutes after the start of the D-glucose infusion. Venous blood glucose levels were measured using a blood gas analyzer (i-Stat, Abbot Scandinavia AB, Sweden). Venous blood glucose curves were not available for subject 7 due to problems with the blood sampling procedure.The subjects were asked to orally report their experiences of the D-glucose infusion after the scanning.

Dynamic acquisition of glucoCEST images was accomplished at a temporal resolution of 36 or 42 seconds, depending on the imaging parameters described in Table 1, and with a total scan time of around 15 minutes. After collecting a number of baseline images (cf. Table 1), D-glucose was administered intravenously and the DGE response was evaluated for selected protocols, all employing a prototype CEST sequence with turbo spin echo (TSE) readout³¹ and a turbo factor of 64. Subjects were imaged using a 3D encoding consisting of eight or ten partitions with a field of view of 212×185 mm², matrix size 64×56, in-plane resolution 3.3×3.3 mm², partition thickness 4 mm and TR/TE=3000/9.0 ms. Saturation was achieved using 10 Gaussian-shaped pulses, either B₁=0.60 μT or B₁=1.6 μT, of 100 ms duration and 10 ms interpulse delay at a single saturation offset of 1.2 ppm or 2.0 ppm from the water resonance frequency, respectively. These offsets are expected to give comparable effects²⁰, with 2.0 ppm being less affected by the steep direct saturation curve and therefore less sensitive to B₀ variations (e.g. due to motion). Two non-saturated (S₀) images were acquired by saturating at +1500 ppm in the beginning of the CEST image acquisition before infusion. An overview of the parameters for the different participants is provided in Table 1.

Table 1.

Overview of glucoCEST acquisition parameters for healthy volunteers with (n=7) and without (n=2) D-glucose infusion.

Subject	Temporal resolution [s]	Number of slices	Saturation offset/B1 [ppm/μT]	Number of dynamics	Baseline: duration [min]	Infusion duration [min]
1-4	42	10	1.2/0.60	25	4	4.0
5	42	10	2.0/1.6	25	4	4.0
6-7	36	8	2.0/1.6	28	6	4.0
A1 #	36	8	1.2/0.60	25	3	-
A2 #	36	8	2.0/1.6	25	3	-
B	36	8	2.0/1.6	18	3	-

^#Subject A was scanned twice, first with 1.2 ppm offset and then with 2.0 ppm offset.

Two of the subjects (denoted A and B) were scanned without D-glucose infusion but with corresponding imaging protocols (cf. Table 1). These subjects did therefore not have PVCs inserted and the venous blood glucose level was not measured. Subject A was scanned twice, with both the 1.2 ppm/0.6 μT and 2.0 ppm/1.6 μT protocols, referred to as A1 and A2, respectively. Subject B was scanned with saturation at 2.0 ppm

In addition to the seven infusion cases scanned according to the description above, eight subjects were infused and scanned using the same pulse sequence but with a single-slice approach instead. As it was not possible to perform a proper retrospective motion correction for these 2D images, the DGE MRI results from these data were omitted. However, four of these subjects had a shorter D-glucose infusion duration of 1.5 minutes, and the venous blood glucose levels and reports of side effects of the infusion are included in the results.

Anatomical T1 MPRAGE images (1.0×1.0×1.0 mm³ isotropic resolution, TI/TE/TR = 900/2.54/1900 ms) were acquired for all subjects. An additional experiment was performed to evaluate possible volumetric changes of the lateral ventricles after D-glucose administration: T1 MPRAGE images were collected in one subject (27 year old male) before and after a 4.0-minute D-glucose infusion (50 mL, 50% dextrose). In total, six image volumes were collected in a total time of 40 minutes. Blood samples were collected at the start of each MPRAGE, and venous blood glucose levels were measured using the same methods as in the glucoCEST experiment described above.

Postprocessing, data analysis and statistics

All data were converted to NIfTI-format and analyzed using Matlab R2019a (The MathWorks, Natick, USA). The first unsaturated (S₀) image and the first saturated image were discarded to assure proper steady state. Retrospective motion correction was performed using the Elastix³² rigid body transform, giving six motion parameters (translation and rotation in each of the three dimensions). A parameter file similar to the one used by Herz et al.²⁵, but with “NumberOfSpatialSamples” set to 2000, was employed. The S₀ image and all the saturated images were registered to the first retained glucoCEST pre-infusion image.

All motion-corrected saturated images were normalized to the motion-corrected S₀ acquired before infusion and the pre-infusion images were averaged to represent a baseline image. ΔDGE maps were generated by subtracting each dynamic glucoCEST image, S(t), from the baseline image, S_base, so that a positive ΔDGE signal can be interpreted as an increased D-glucose concentration in the voxel ²:

$$\Delta \text{DGE}(\mathrm{t})[\%]=\frac{S_{base}-S(t)}{S_0}\cdot 100\%.$$ (1)

Averaged area under curve (AUC_mean) maps were created as following:

$${\text{AUC}}_{\text{mean}}[\%]=\frac{{\sum }_{t_1}^{t_2}\Delta \text{DGE}(\mathrm{t})}{N},$$ (2)

where N is the number of images over a selected time interval Δt = t2 – t1. The division by N was performed to achieve comparable normalized signal magnitudes regardless of temporal resolution. AUC_mean maps were calculated for each subject in four two-minute time intervals. The first interval (referred to as baseline) corresponded to the initial two minutes of the period before D-glucose infusion. The second, third and fourth intervals represented post-infusion intervals, starting 2, 4 and 6 minutes after the start of the D-glucose infusion, respectively. The AUC_mean time intervals are schematically illustrated in Figure 1. A baseline of 3 minutes were assumed in the AUC_mean calculations for the subjects without infusion, and four 2-minute AUC periods were analyzed similar to the actual infusions.

Figure 1.

Schematic illustration showing the different intervals of the DGE data acquisition and analysis. Each image series started with two S₀ images, followed by a baseline period of varying duration, infusion and post-infusion periods. For the AUC data analysis, 2 min segments were used, one representing the first two minutes of the baseline and three representing post-infusion intervals, starting 2, 4 and 6 minutes after the start of the D-glucose infusion.

Anatomical MPRAGE images were resliced to correspond to the glucoCEST images in terms of spatial resolution and slice positioning, using SPM12³³. ΔDGE time curves were measured in regions of interests (ROIs) of a cerebral artery, in white matter (WM), and in cerebrospinal fluid (CSF) in the lateral ventricles, identified in the resliced anatomical images. The ΔDGE time curves were temporally smoothed using a 3-point moving mean (Matlab function movmean). The AUC_mean values for the four 2-minute intervals described above were calculated for the ROIs in artery, WM and CSF. All subjects that received D-glucose infusion (n=7) were grouped together and the ΔDGE AUC signal in the arterial ROIs after D-glucose infusion in the second to fourth interval, compared to the baseline, was evaluated using two-sided Wilcoxon signed-rank tests.

To further evaluate the effect of motion on the ΔDGE signal, rigid body motion estimates (three degrees of translations and rotations for each volume in time) were applied to the first retained glucoCEST pre-infusion image to simulate an artificial time series. This procedure was executed in Matlab with the imwarp function using a rigid body 3D transformation. A pseudo-DGE time series where the ΔDGE signal was entirely caused by movements was then calculated using equation 1. The average ΔDGE signal in the imaged volume over time was calculated for each subject before and after motion correction as well as in the pseudo-DGE maps. The outer two slices were excluded from the analysis, leaving 6 or 8 slices depending on the protocol. This exclusion was necessary in order to avoid erroneous signal contributions caused by the motion correction in the outermost slices.

For the analysis of the ventricle volume, the lateral ventricles were segmented out using FSL FAST segmentation³⁴ followed by manual delineation, and the volume of the lateral ventricles were calculated at each time point. The two volume measurements before D-glucose infusion were averaged to represent a baseline value. The change in ventricular volume was plotted as a function of time together with venous blood glucose levels. To evaluate the quality of the ventricular volume estimation, an additional scan of the same subject, but without D-glucose infusion, was performed in a separate session three weeks before the glucose infusion scan. Three MPRAGE volumes were collected and the volume of the lateral ventricles was calculated in each of these image volumes as described above.

A two-sample t-test was used to compare the maximum change in blood glucose level between the 1.5 min and 4.0 min infusion groups, including subjects from both the single-slice and the 3D acquisitions. The baseline venous blood glucose level (BGL) was subtracted to obtain the change in blood glucose level. The relationship between DGE AIFs and venous blood glucose levels was evaluated using a linear regression analysis.

An additional analysis of B₀ shifts and their effect on the DGE signal for 1.2 ppm/0.6 μT and 2.0 ppm/1.6 μT was performed, and is described in detail in the Supplementary Material.

Results

Motion-corrected DGE AUC_mean maps of one slice for three representative subjects are shown in Figure 2A. Four AUC_mean time intervals of two minutes each are shown. A signal change after D-glucose infusion was seen in all subjects, mainly manifested as hypointensities. Motion artifacts (residual uncorrected motion, seen as dark-bright patterns) appeared at tissue interfaces in some subjects, most prominent around the ventricles and in outer parts of the brain. One subject (Subject B) without infusion is also included in Figure 2 and a signal change can be seen, especially near the edges of the brain and and in the later AUC_mean intervals. When compared visually, the overall signal change was smaller for the subjects without infusion than in the D-glucose infusion group. Figure 2B includes dynamic curves for ROIs in an artery, WM and CSF presented together with the change in venous blood glucose level (ΔBGL). An increase in the arterial ΔDGE signal of approximately 1-4% following the D-glucose infusion was observed in all subjects. The ΔDGE signal in white matter was small in most subjects, compared to the arterial signal, and in many cases it was negative. The ΔDGE signal in CSF was in some cases large (2-4% in subjects 1, 4 and 6), but positive or negative. In the group without infusion, the ΔDGE signal was close to zero in all tissues, except for subject A1, where a fluctuating ΔDGE signal of ±1% could be seen in artery and CSF. A scatter plot between the DGE AIFs and the corresponding change in venous blood glucose level for all subjects that had D-glucose infusion, except for subject 7, is presented in Figure 2C. A significant correlation of r=0.65 was achieved.

Figure 2.

(A) ΔDGE AUC_mean maps of two representative subjects at 2-minute intervals taken before and after D-glucose infusion (cf. Figure 1). The third subject, subject B, was imaged without D-glucose infusion. (B) ΔDGE signal in artery, white matter and cerebrospinal fluid together with the change in venous blood glucose level (ΔBGL) as a function of time for each subject. The regions of interest (ROIs) in which the ΔDGE signals were measured are indicated in the anatomical images in each plot. The time zero represents the start of the D-glucose infusion and the shaded areas in the graphs represent the infusion duration. (C) Scatter plot of arterial ΔDGE signal in one slice at all time points when blood glucose was measured and corresponding change in venous blood glucose level for all subjects that had D-glucose infusion and successful blood glucose sampling (1.2 ppm, n=4; 2.0 ppm, n=2).

Table 2 shows the maximum displacement (translation and rotation) retrieved from the rigid body motion estimates for each subject. Subjects 1, 4 and 6 had remarkably higher motion estimates than the other subjects, corresponding to translations of 1.7-2.1 mm and rotations of 0.8-1.9 degrees. Individual correlation coefficients between the change in venous blood glucose level and the corresponding arterial ΔDGE signal for subjects that had D-glucose infusion are presented in Table 2. The correlations were all positive, but only two out of six were significant. Correlation coefficients between the change in venous blood glucose levels and ΔDGE signal in white matter are also included in Table 2. Negative significant correlations were found in three subjects.

Table 2.

Maximum movements (translation and rotation) retrieved from the motion correction, and correlation coefficients (r) and p-values (p) between ΔDGE in artery and venous blood glucose levels (BGLs), and between ΔDGE in white matter and ΔBGL for all subjects.

	Maximum Displacement		Correlation Venous BGL – Artery		Correlation Venous BGL – WM
	Translation [mm]	Rotation [degrees]	r	p	r	p
1	2.1	1.9	0.48	0.3	−0.83	*
2	0.9	0.9	0.52	0.2	−0.72	*
3	0.5	0.5	0.57	0.1	−0.32	0.4
4	1.7	1.8	0.97	*	−0.46	0.3
5	0.6	0.3	0.54	0.2	0.20	0.7
6	1.7	0.8	0.81	*	−0.86	*
7	0.6	0.3	N/A	N/A	N/A	N/A
A1 #	0.4	0.4	N/A	N/A	N/A	N/A
A2 #	0.3	0.3	N/A	N/A	N/A	N/A
B	0.1	0.6	N/A	N/A	N/A	N/A

Venous blood D-glucose levels were not available for one subject due to blood sampling failure (subject 7). The asterisk (*) indicates p < 0.05. N/A = not available. Shading is used to help distinguish between the different protocols; subjects 1-4, (1.2 ppm, 4.0-minute infusion), subjects 5-7 (2.0 ppm, 4.0-minute infusion) and subjects A-B (no infusion).

^#Subject A was scanned twice, first with 1.2 ppm offset and then with 2.0 ppm offset.

Figure 3 shows the ΔDGE signal in the selected regions in artery, WM and CSF of all subjects, grouped according to the following: 1.2 ppm/0.6 μT with infusion (n=4) and without infusion (n=1) as well as 2.0 ppm/1.6μT with infusion (n=3) and without infusion (n=2). The last two rows of Figure 3 show the same data but instead grouped into one group with infusion (n=7) and one group without infusion (n=3). All ΔDGE data points in each time interval are included for each subject. When grouping the subjects with infusion together (n=7), a significant signal increase in ΔDGE signal in arteries compared to baseline was found in all post-infusion time intervals (p=0.03 for all intervals).

Figure 3.

Box-and-whisker plots showing ΔDGE signal in arterial, white matter and ventricular regions of interest grouped into four different two-minute intervals. The time intervals were as follows: Baseline, representing the initial two minutes of imaging, and three post-infusion intervals starting 2, 4 and 6 minutes, respectively, after the start of D-glucose infusion. The red line represents the median value, the upper and lower borders of the blue box represent the upper and lower quartile, respectively, and the end points of the whiskers represent the maximum and minimum DGE value in each group. The bottom two rows show the grouped results for the two frequency offsets.

In Figure 4, pseudo-DGE maps (DGE_Pseudo) are shown together with real non-corrected DGE (DGE_NonCo) and motion-corrected DGE (DGE_MoCo) maps for all subjects, using the AUC_mean for the 2-4 minutes period post infusion (AUC_{2-4 min}). DGE_Pseudo maps and DGE_NonCo maps were similar but not identical when compared visually. Figure 5 shows the ΔDGE average over the imaged volume (outer slices excluded) as a function of time, together with the motion estimates (translations and rotations) for five subjects. The averaged ΔDGE signal was in some cases altered after motion correction, as seen in subjects 1 and 6.

Figure 4.

ΔDGE AUC_{2-4 min} maps of all subjects. Three different ΔDGE map versions are included for each subject: non-corrected (NonCo), pseudo-, and motion-corrected (MoCo) DGE maps. The pseudo-DGE maps were created by applying the motion estimates to the first baseline image, creating artificial ΔDGE time series where the contrast change is caused entirely by motion.

Figure 5.

Averaged ΔDGE signal over the imaged volume (outer slices excluded) as a function of time calculated in the non-corrected (NonCo), pseudo-, and motion-corrected (MoCo) DGE maps, shown together with the corresponding motion estimates (translations and rotations in the x-, y-, and z-direction) for five subjects. The time zero represents the start of the D-glucose infusion and the shaded areas represents the infusion duration.

Figure 6 shows DGE_NonCo, DGE_Pseudo and DGE_MoCo time series of a center slice for the subject scanned without D-glucose infusion using both saturation offset protocols (1.2 ppm/0.6 μT and 2.0 ppm/1.6 μT, referred to as subject A1 and A2, respectively). The 2.0 ppm images have a smoother appearance after motion correction than the 1.2 ppm images.

Figure 6.

A subject scanned without D-glucose infusion, first with the 1.2 ppm/0.6 μT saturation protocol (left) and then with the 2.0 ppm/1.6 μT saturation protocol (right). Three different ΔDGE time series are shown for each saturation frequency offset; top row: non-corrected DGE maps, middle row: pseudo-DGE maps and bottom row: motion-corrected DGE maps.

As stated in the methods section, DGE data from the single-slice acquisitions were not included. The results from the D-glucose infusion in these subjects are, however, included in the following section. The average baseline blood glucose level ±SD was 4.9±0.9 mM and the average of the maximum change in blood glucose levels were 8.6±2.3 and 8.7±1.8 mM for the 1.5- and 4.0-minute infusion groups, respectively. No significant difference in maximum change in blood glucose level was found between the 1.5-minute (n=4) and the 4.0-minute (n=12) infusion groups (p=0.96). The verbally reported D-glucose side effects of the 1.5-minute infusion (n=4) were a warm or cold feeling or tension at the injection site (100%, all subjects), a warm or pulsating sensation in the head and/or crotch (75%, three subjects), transient headache (75%, three subjects), experiencing an urge to urinate (50%, two subjects), and tension in the shoulder (25%, one subject). The reported side effects in the 4.0-minute infusion group (n=12, including the ventricle scan) were a warm or cold feeling or tension at the injection site (67%, eight subjects) and a warm or pulsating sensation in the head and/or crotch (25%, three subjects). The rest of the subjects (33%, four subjects) did not experience any side effects. One subject in the short infusion group reported a thrombophlebitis close to the site of injection (the basilic vein) a few days after the examination. Evaluation of the morphological images did not reveal any brain pathology in any of the examined subjects.

Segmentation of the lateral ventricles, without D-glucose infusion, resulted in a volume of 22.9 mL. The difference between the first and second image volume and between the first and third image volume was <0.1%. Results of the measurement with D-glucose infusion on the same subject are presented in Figure 7. The averaged baseline volume was 22.3 mL and the average volume after infusion was 22.5 mL, which corresponds to an average change of about 1%. This subject did not experience any side effects of the D-glucose infusion.

Figure 7.

Measurement of the volumetric changes of the lateral ventricles in one subject before and after D-glucose infusion shown together with the measured venous blood glucose level. The image in the lower right shows one of the segmented lateral ventricle volumes.

Results of the analysis of B₀ shifts and their effect on the DGE signal are provided in the Supplementary Material.

Discussion

DGE AUC_mean images and protocol comparison

Signal changes were seen in all DGE AUC_mean maps after D-glucose injection. Of the nine subjects, five were scanned using saturation at 1.2 ppm, a frequency often utilized at higher field strengths (i.e. 7 T and above), because three of the D-glucose hydroxyl protons are resonating at around 1.2 ppm. However, these three protons have an exchange rate of thousands of Hz¹⁸ and, are coalesced with the water proton resonance at 3 T, leading to a steep slope in the Z-spectrum at this frequency. Previously, based on simulations, it was therefore recommended to use a saturation frequency at 2.0 ppm²⁰ because the sensitivity to B₀ inhomogeneities decreases when saturating farther from the water resonance frequency^18,23,35. As shown by Zaiss et al.²³, B₀ induced frequency shifts caused by subject motion or field drift can manifest themselves as global hypo- or hyperintensities in DGE maps. No dynamic B₀ correction was performed in this study, because a high temporal resolution was prioritized over multiple frequencies to enable measuring AIFs, which was accomplished by using a single saturation offset. Subject A was scanned without D-glucose infusion using both saturation protocols, and the motion estimates were of comparable magnitude. The 1.2 ppm ΔDGE maps showed more signal fluctuations than the 2.0 ppm ΔDGE maps, as seen in Figure 6. This illustrates that the 1.2 ppm images are more affected by motion-induced B₀ artefacts. The effect of B₀ shifts on the DGE signal for 1.2 ppm/0.6 μT and 2.0 ppm/1.6 μT was further investigated (see Supplementary Material). As reported by others ^23,36, this effect can be substantial. For instance, our analysis revealed that a typical shift of 1 Hz at 3 T can change the absolute size of the AIF by as much as 0.2 percentage points. In WM, this would be around 0.2%/Hz for 1.2 ppm/0.6 μT and around 0.1%/Hz for 2.0 ppm/1.6 μT. These results indicate that single saturation offset acquisition of glucoCEST images can be problematic, especially at saturation offsets closer to the water resonance. Hence, it appears that the use of single frequencies for DGE acquisition should be replaced by acquisition of the Z-spectrum, at least including a region around 0 ppm to estimate and correct for the field shifts due to motion or drift.

Hypointense areas over the whole brain were present in the AUC_mean maps (Figure 2) for both the 1.2 ppm/0.6 μT and 2.0 ppm/1.6 μT groups. Generally, the AUC_mean in WM was slightly negative for all the post-infusion intervals (Figure 3). This observation is consistent with earlier results showing a reduced D-glucose effect in white matter for a prolonged period of 10-20 minutes^20,25, but it has not been entirely explained. According to Equation 1, a positive ΔDGE signal would be expected in regions where the D-glucose concentration has increased compared to baseline. There is no physiological explanation to why the tissue glucose concentration should decrease directly after D-glucose injection, so the source behind the negative ΔDGE signal seen in some regions and subjects must be an MR-related phenomenon (i.e. water signal related). Due to the consistency of the reduction, it seems unlikely to be caused by B₀ shifts due to motion. Another possible source could be B₀ drift caused by gradient heating. However, the negative signal in WM tends to recover towards the end of the ΔDGE time series, as seen in the time curves in Figure 2, which suggests that it is related to the change in blood glucose concentration. A significant negative correlation between ΔDGE signal in WM and venous blood glucose level was found in 50% of the subjects (cf. Table 2). Possible causes for this decrease may be a susceptibility-based frequency shift or a tissue water signal change due to osmolarity differences between blood and tissue that is larger than the glucoCEST effect at 3 T but smaller than the corresponding effect at 7 T²⁰. As seen in Figure 3, the ΔDGE signal in WM was typically close to zero or positive in the subjects without D-glucose infusion. Additionally, the averaged ΔDGE signal over the imaged volume, shown in Figure 6, is typically close to zero for DGE_Pseudo, which is expected since motion-induced signal changes are likely to cancel out. This behavior is not observed in the DGE_MoCo, showing that motion is not the dominating signal contributor in the motion-corrected DGE maps. Even though the cases without infusion are few, they are important as a reference and illustrate the difficulties in DGE imaging, and that careful post-processing and analysis are needed to distinguish true ΔDGE signal from ΔDGE signal caused by motion.

Negative and positive signal changes were found in CSF, as exemplified here in the lateral ventricles. The DGE_Pseudo maps (Figure 4) reveal that motion can cause hypo- or hyperintensity in and around the lateral ventricles. No consistent trend is seen for CSF in Figure 3, indicating that at least part of the signal change in CSF can be attributed to motion. A complementary explanation is volumetric changes of the ventricles after D-glucose injection, causing a water signal decrease due to partial volume effects with neighboring tissues interpreted as an apparent decrease in D-glucose concentration². This can also explain why the signal change in the ventricular region is smaller after rigid motion correction in the cases without infusion than in the D-glucose infusion group, since the former only includes rigid motion while the latter also consists of non-rigid motion due to ventricular swelling or shrinking. A volumetric change of up to 2% was measured in our experiment performed on one volunteer, as seen in Figure 7. When looking at the results, there are three points with a volume change of 1-2 %, but one with a negative change, reducing the average. The cause for this deviating data point is probably motion related, as this fourth image was, by visual inspection, blurrier at tissue interfaces than the others. Even though this would be an unbiased criterion for not including it, we felt it was appropriate to show all data. In support of the overall results, a previous study²⁷ by Puri et al. observed a similar volume increase of the lateral ventricles of 2.4±0.4% after raising the blood glucose level in healthy volunteers from 4.8±0.2 mM to 8.4±0.4 mM using oral administration of D-glucose.

Relationship between arterial DGE signal and blood glucose levels

DGE AIFs have previously been measured in healthy volunteers at 7 T¹³ and in patients at 3 T²⁰. In this study at 3 T, a significant increase in the averaged arterial ΔDGE signal was found in the group with D-glucose infusion (n=7), and we thus conclude that it is possible to measure arterial ΔDGE signal at 3 T. One limitation of this study is the small number of subjects that have been investigated, and this limitation is only partially overcome by grouping all the infusion subjects with different acquisition conditions. The magnitude and shape of the arterial ΔDGE curve varied between individuals. The magnitude differences can most likely be explained at least in part by partial volume effects with surrounding tissue. A reasonable explanation for the finding of different magnitude combined with different shape of the DGE AIFs between subjects is the individual response to the D-glucose in terms of insulin response and metabolism, which will influence the curve shape as discussed by Knutsson et al.¹³. The subjects scanned using saturation at 1.2 ppm generally showed more fluctuations in the ΔDGE curves, probably because this is more sensitive to frequency shifts caused by changes in B₀, as discussed in the previous section.

D-glucose infusion duration, venous blood glucose levels and D-glucose side effects

The blood glucose sampling procedure is challenging. Long plastic (polycarbonate) tubing is needed to avoid the nurses leaning into the magnet bore when collecting the blood samples. Tubing of 100 cm length was used in our previous glucoCEST experiments to facilitate the blood sampling process, but this was changed to a length of 25 cm. For one of the subjects, subject 7, it was not possible to retrieve blood through the tubing and the samples were thus limited to one sample before and one sample taken at the arm after the scanning. Saline flushes are necessary in order to assure that fresh blood is retrieved at each sampling time. Due to the blood sampling setup, there was a small risk that the extracted blood got contaminated by saline and a non-reliable blood glucose level would then be measured. Such cases were confirmed by the blood gas analyzer showing unreasonably small values, so these points could safely be excluded. When DGE measurements, at some point, becomes standardized, continued blood sampling will be unnecessary and such side effects and related motion effects would be eliminated.

In this study, a D-glucose infusion with commercially available 50% dextrose was used. Fast infusion of D-glucose in high concentration solutions can traumatize the vein due to its high osmolarity³⁷. Thrombophlebitis was reported by one of our participants and has also been documented for one volunteer in another glucoCEST study ²⁶.

In a study by Chen and Porte²⁹, the insulin and blood D-glucose responses to different D-glucose injection rates were investigated. Using three different injection durations (0.3, 3 and 12 minutes), they showed that a faster infusion gave a steeper rise in blood glucose level without significantly affecting the peak level. A higher D-glucose infusion rate was associated with a faster disappearance of D-glucose from the circulation and a higher acute insulin response. In our study, it was difficult to compare the shape of the sampled venous blood glucose curves between participants since the sampling times sometimes varied between subjects. However, no significant difference in maximum blood glucose level change (baseline value subtracted) was found (p=0.96) between the 1.5-minute and the 4.0-minute infusion groups. Using these results together with the results from the study by Chen and Porte²⁹, we conclude that a longer infusion duration (e.g. 3-4 minutes) is preferable in glucoCEST experiments, since it could minimize the side effects of the D-glucose infusion while not reducing the maximum blood glucose level and therefore preserving the DGE effect size. A longer infusion duration can thus reduce the risk of subject motion during and immediately after D-glucose infusion. In this study, all subjects were asked about their experiences during the D-glucose infusion. We observed that a longer infusion duration was experienced as more pleasant for the participant, and this is in agreement with previous observations¹³.

Conclusions

DGE imaging at 3 T remains challenging and the need for thorough post-processing and analysis is emphasized, including motion correction and dynamic B₀ correction, which would be possible by using multiple saturation frequencies in the acquisition. Movements of the order of 2 mm and 1.5 degrees obscure the ΔDGE signal and can lead to large signal changes in CSF. In spite of the challenges, we conclude that it is possible to measure arterial ΔDGE signal at 3 T. A longer infusion duration (e.g. 3-4 minutes) should preferably be used in glucoCEST experiments, since it can minimize the side effects of the D-glucose infusion while still producing sufficient change in blood glucose level. Future studies on larger cohorts are needed to address post-processing, motion correction, dynamic B₀ correction, physiological changes related to D-glucose infusion, as well as, most relevantly, new pulse sequences to increase the effect size.

Abbreviations used:

AIF

arterial input function

AUC

area under curve

BGL

blood glucose level

CEST

chemical exchange saturation transfer

CSF

cerebrospinal fluid

DGE

dynamic glucose-enhanced

DSC

dynamic susceptibility contrast

MTR

magnetization transfer ratio

PVC

peripheral venous catheter

ROI

region of interest

WM

white matter

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.