FDG uptake of the normal spinal cord in PET/MR imaging

Background: The lack of visualization of the spinal cord hinders the evaluation of 18 F Fluoro-deoxy-glucose (FDG) uptake of the spinal cord in PET/CT. By exploiting the capability of MRI to precisely outline the spinal cord, we performed a retrospective study aimed to define normal pattern of spinal cord FDG uptake in PET/MRI. Methods: Forty-one patients with lymphoma without clinical or MRI signs of spinal cord or bone marrow involvement underwent simultaneous PET and MRI acquisition using Siemens Biograph mMR after injection of 3.5 MBq/kg body weight of 18 FDG for staging purposes. Using a custom-made software, we placed ROIs of 3 and 9 mm in diameter in spinal cord, lumbar CSF and vertebral marrow that were identified on MRI at 5 levels (C2, C5, T6, T12 and L3). The SUVmax, SUVmean and the SUVmax and SUVmean normalized (NSUVmax and NSUVmean) to the liver were measured. For comparison, the same ROIs were placed in PET-CT images obtained immediately before the PET-MRI acquisition following the same tracer injection. Results: On PET/MRI using the 3 mm ROI the following average (all level excluding L3) spinal cord median (1st and 3rd quartile) values were measured: SUVmean1.68 (1.39 and 1.83), SUVmax1.92 (1.60 and 2.14), NSUVmean1.18 (0.93 and 1.36), NSUVmax1.27 (1.01 and 1.33). Using the 9 mm ROI the corresponding values were: SUVmean1.41 (1.25-1.55), SUVmax2.41 (2.08 and2.61), NSUVmean0.93 (0.79 and 1.04), NSUVmax1.28 (1.02 and 1.39). Using the 3 mm ROI the highest values of PET-MRI SUVmax, SUVmean, NSUVmax and NSUVmean were consistently observed at C5 and the lowest at T6. Using a 9 mm ROI the highest values were consistently observed at C5 and the lowest at T12 or T6. The spinal cord FDG-uptake values correlated with bone marrow uptake at the same level, especially in case of NSUVmax. Comparison with PET-CT data revealed that the average SUVmax and SUVmean of the spinal cord were similar in PET-MRI and PET-CT. However, the average NSUVmax and NSUVmean of the spinal cord were higher (range 21% - 47%) in PET-MRI than in PET-CT. Conclusions: Using a whole-body protocol we defined the maximum and mean FDG uptake of the normal spinal cord in PET/MRI. While the observed values show the expected longitudinal distribution, they appear to be higher than those measured in PET/CT. Normalizing the SUVmax and SUVmean of the spinal cord to the liver radiotracer uptake could help in multi-institutional comparisons and studies.

Conclusions: Using a whole-body protocol we defined the maximum and mean FDG uptake of the normal spinal cord in PET/MRI. While the observed values show the expected longitudinal distribution, they appear to be higher than those measured in PET/CT. Normalizing the SUVmax and SUVmean of the spinal cord to the liver radiotracer uptake could help in multi-institutional comparisons and studies.

Background
Functional imaging of the spinal cord with positron emission tomography (PET) represents a distinctive challenge due to the its small size, the relatively low resolution of PET images and contamination from vertebral bone marrow [1]. Studies published so far utilized PET/CT and the metabolic tracer 18 F Fluoro-deoxy-glucose (FDG) according to a whole-body [2,3] or dedicated spinal protocol [4].
The more reliable measurement of FDG uptake in PET/CT studies of normal and diseased spinal cord is uncertain [1]. Two approaches have been utilized. One entails measurement of the standardized uptake value (SUV) of the normal or affected spinal cord per se that is usually sampled with a Region Of Interest (ROI) encompassing the entire spinal canal due to lack of the spinal cord definition and taking care to exclude the vertebral bone. The other implies measurement of the ratio of SUV to background using the activity of the normal spinal cord, lumbar thecal sac, vertebral marrow, or liver as a reference [2,3,4]. Moreover, due to inclusion of the surrounding CSF in the spinal cord ROI sampling in PET/CT, the spinal cord SUVmax, that is assumed to reflect the higher metabolic activity of the cord central gray matter, was usually preferred to SUVmean in PET/CT studies of spinal cord diseases.
To tackle the problem of lack of visualization of the spinal cord, which hinders evaluation of FDG uptake of the cord in PET/CT, it is now possible to take advantage of the introduction of hybrid PET/MR scanners. In fact, they allow to simultaneously acquire both metabolic (through PET imaging) and morphological data with excellent tissue contrast (via MRI) [5,6]. Moreover, simultaneous hybrid systems can also overcome the problem of the PET and MR coregistration, since PET and MR images (of the same anatomical district) are obtained at the same time and can therefore ideally coregistered [7]. Moreover, the visualization of the spinal cord contour in PET/MRI enables measurement of the SUV across the entire ROI (SUVmean) that can represent an additional useful tool in certain disease conditions. The aim of this study was to demonstrate the feasibility of PET/MR imaging of the spinal cord. In particular, by exploiting the capability of MRI to precisely outline the spinal cord, we performed a semi-automated ROI analysis of the FDG uptake in PET/MRI in adults without evidence of spinal cord or bone marrow abnormalities. Our results were compared to the available data on PET/CT imaging of the spinal cord and aimed to define the normal pattern of spinal cord FDG uptake in PET/MRI.

Material And Methods
This is a retrospective study performed in a single center where two hybrid PET scanners, one PET/CT and one PET/MRI, operate. Scan were performed after 6 hours of fasting. Body weight, height, and glucose levels were measured. After intravenous injection of a dose 3.5 MBq of FDG/kg body weight, patients were resting for 45 min. During whole body acquisition the patient was placed supine in the scanner so that he or she was comfortable, and motion could be minimized during the acquisition. Patients were examined sequentially with PET/CT (10 minutes duration) and PET/MRI (60 minutes duration) exploiting a single injection of radiotracer, thus avoiding delivery of additional ionizing radiations as compared to PET/CT alone. The patient provided a written informed consent before the injection and acquisition.

Patients selection
Forty-seven patients with early stage lymphoma prior to treatment and not in an hyperacute phase were retrospectively enrolled. The following exclusion criteria were applied: (1) presence of motion between PET and CT datasets with failed/poor coregistration (5 patients were excluded); (2) known spinal canal or cord diseases based on existing imaging and chart review (1 patient with intramedullary tumor was excluded). Finally, 41 patients were selected (20 men and 21 women) with a mean age of 46.9±18.7 years.

FDG PET/CT acquisition
A Discovery IQ hybrid (GE Healthcare, Milwaukee, WI, USA) PET-CT scanner was used with tube voltage, 140 kVp; pitch, 0.94; reconstructed slice thickness, 3.75 mm; increment, 3.26 mm. PET data were acquired using the three-dimensional (3D) mode with a fixed scan duration of 2 minutes per bed position. Emission data were corrected for random coincidences, dead time and scatter. CT data were used for attenuation correction and assistance in anatomic localization of FDG. PET emission data were reconstructed with ordered subsetexpectation maximization (OSEM) algorithm (21 subsets, 4 iterations) and post-filtered with a three-dimensional isotropic gaussian of 4 mm full width at half maximum.

FDG PET/MRI acquisition
A 3T mMR Siemens Biograph hybrid PET-MR system (Erlangen, Germany) was used [8], with PET images (4 mm spatial resolution) acquired simultaneously (5 minutes per bed position) with whole body MRI T2 HASTE sequence (TR, 1400; TE, 87; slice thickness, 6mm; field of view, 380 x 380) and automatically corrected for tissue attenuation with the attenuation maps generated by using Dixon MRI sequence (Pace et al, 2013)(first TE, 1.23 ms; second TE, 2.46 ms; TR, 3.96 ms; voxel size, 4.1 × 2.6 × 3.1 mm; field of view, 500 × 312). Emission data were corrected for random coincidences, dead time, scatter, and attenuation. PET data were reconstructed with a 3D attenuation weighted ordered-subsets expectation maximization iterative reconstruction algorithm (AW OSEM 3D) with three iterations and 21 subsets, gaussian smoothing 4 mm full width at half maximum.

ROI and statistical analyses
We used a custom-made software, named SMART anatomical LABelling (SMARTLAB) tool for medical imaging datasets, based on the Syngo.Frontier platform (Siemens, Germany), to place fiducial points in PET/MRI and PET/CT images. Fiducial points in spinal cord, lumbar CSF and vertebral marrow were identified on MRI and CT at five levels (C2, C5, T6, T12 and L3). An additional fiducial point was placed in the liver, on the 6th hepatic segment [9], avoiding blood vessels. The space containing the fiducial point was imported on a MATLAB preprocessing pipeline which includes the re-slicing and co-registering of PET/MRI and PET/CT images on the space of the fiducial points. From the fiducial point a circular ROI of 3mm radius and a spherical Volume of Interest (VOI) of 9mm radius was calculated for each level. The ROI/VOI were centered on the spinal cord and the corresponding vertebral body bone marrow ( Figure 1). Then the SUVmax and the SUVmean were measured. In addition, as suggested by Marini et al. [4], a reference ROI of in the liver was measured in both PET/MRI and PET/CT and enabled to calculate the Normalized SUVmax (NSUVmax) and the Normalized SUVmean (NSUVmean).
A preliminary Shapiro-Wilk test showed that the data distribution was not normal. Accordingly, we adopted nonparametric tests for statistical analyses. We used Wilcoxon test to compare spinal cord and bone marrow FDG uptake in PET/MRI and PET/CT and the Spearman rank correlation test to evaluate the correlation between spinal cord and bone marrow uptake at the same spinal level. Bonferroni correction for multiple comparison was applied (p-value <0.01).

Results
Only the NSUVmax of the spinal cord in 3 or 9mm ROI/VOIs was significantly different between PET/MRI and PET/CT and consistently higher in the former ( Table 1). The SUVmax, SUVmean and NSUVmean in spinal cord are reported in supplementary materials (Table S1).
A significant difference between PET/MRI and PET/CT was also observed for the SUVmax and NSUVmax in bone marrow using both 3mm and 9mm ROI/VOI with consistently higher values in the former ( Table 2). SUVmean and NSUVmean in bone marrow are reported in supplementary materials (Table S2).   Figure 2 shows the distribution of the spinal cord NSUVmax along the vertebral levels in PET/MRI.
In PET-MRI the spinal cord FDG-uptake values correlated with bone marrow uptake at the same level, especially in case of NSUVmax (table 3).  Table 4 shows the comparison of the average SUVmax and SUVmean and NSUVmax and NSUVmean of the spinal cord, CSF at L3 and average vertebral bone marrow in PET/MRI and PET/CT in our patients. In general, they were similar, but the average NSUVmax and NSUVmean of the spinal cord were higher (range 21%-47%) in PET/MRI than in PET/CT. On the other hand, the SUVmax and SUVmean of CSF at L3 was lower in PET/MRI than in PET/CT.

Discussion
So far the FDG uptake in normal and diseased spinal cord was evaluated in few PET/CT studies [2,3,4,10,11,12,13,14]. It is recognized that clinical PET/MRI could be advantageous with respect to PET/CT for assessment of spinal cord diseases [1]. PET/CT studies have shown that gender, age, body mass index and blood glucose have no significant impact on SUVmax of the spinal cord [3,10,11,14]. However, in pediatric population FDG uptake of the spinal cord increased with body weight [14].
We are aware of several technical and methodological differences when the average SUV values we measured in PET/MRI are compared with those measured in PET/CT by us (and others). We administered an FDG dose of 3.5 MBq/kg body weight that is lower to those used in prior FDG/PET studies (4.8 -5.2 MBq/kg of body weight [4]). However, while a lower signal to noise ratio can result from lower doses, no effect on SUV is expected [15]. In our protocol we subsequently acquired PET/CT and PET/MRI. While this may imply lower PET values in PET/MRI, the duration of acquisition was reported not to affect SUVmax values [16].
Our method for ROI computation of the spinal cord FDG uptake values in PET/CT was different from that employed in prior PET/CT studies. In fact, we coregistered in PET/CT the ROIs that we drew on MRI taking advantage of the spinal cord direct visualization. In prior PET/CT studies the spinal cord uptake refers to the SUVmax of a ROI that is drawn along the contour of the spinal canal with care to avoid inclusion of the cortical vertebral bone. This notwithstanding, Patel et al. [3] emphasized the need of correcting for bone marrow contamination when reporting FDG uptake of the spinal cord. Therefore, our spinal cord ROI annotation procedure could be more accurate with respect to prior studies.
Once all the above different features are considered, it is noteworthy that the SUVmax values of the normal spinal cord we measured in PET/CT as at each spinal level or on average over the entire spinal cord were similar or slightly higher than those measured in other PET/CT studies [3,4,10,11,14], especially when we used the 9mm VOI. This presumably reflects the lower contamination from CSF in the perimedullary space and from vertebral marrow in our ROI placement procedure. This notwithstanding, we measured higher average FDG uptake values (max and mean) of the spinal cord in PET/MRI than in PET/CT.
Moreover, it is conceivable that if a direct PET/MRI acquisition could have been performed without prior PET/CT, FDG uptake values of the spinal cord in PET/MRI in our patients could even be higher than those we measured. This is important since it may imply a wider dynamic range of uptake values of potential use in PET-MRI assessment of disease conditions of the spinal cord.
Remarkably, the longitudinal distribution of SUVmax and NSUVmax of the spinal cord we measured in PET/MRI is in line with that previously reported in PET/CT [2,3,10,14]. In particular, using 3 mm ROI we observed the highest SUVmax values at vertebral level C5 and the lowest values at vertebral level T6. This feature has been attributed to the variable amount of nervous tissue and in particular of the metabolically more active central gray matter in the spinal cord [2]. In fact the cross-sectional area of the spinal cord decreases from C1 to the conus medullaris with the exception of cervical (from C3 to T2 vertebral body) and lumbar (from the T9 to T12) enlargements, where the transverse diameter increases due to the relative expansion of the gray and white matter that is associated with a greater number of neurons correlated to upper and lower limbs sensory and motor functions. In particular, the C4 to T1 spinal cord neural metamers, corresponding to the vertebral body level from C3 to T2, show a mean cross-sectional area of gray matter from 7.8 to 10.7 mm 2 and the L3 to S1 spinal cord neural metamers corresponding to vertebral body level from T9-T10 to L1-L2 show a mean cross sectional of gray matter of 13.2 to 16.7 mm. Differently, the remainder thoracic spinal cord neural metamers show a range from 3.7 to 5.6 mm 2 of mean cross section of the gray matter [17,18]. Accordingly, the SUVmax of the FDG uptake of the spinal cord would reflect different segmental (metameric) levels of specialization and demands with higher values in segments involved in sensory and motion functions of the limbs and lower values in thoracic segments mainly involved in the functions of thoracoabdominal visceral organs and sensory and motion functions of the trunk.
The significant correlation between spinal cord and bone marrow FDG uptake values we observed in both PET/MRI and PET/CT is in line with the PET/CT data by Patel et al. [3]. A distinct advantage provided by PET/MRI as compared to PET/CT is the possibility of reliably measuring the SUVmean, namely the mean FDG uptake of the cross-sectional area of the spinal cord that is not visible in PET/CT. As expected, inclusion of the cord white matter with its lower metabolism/glucose consumption in the ROI implies lower SUVmean values as compared to the SUVmax values which essentially reflects the higher metabolism/glucose consumption of the central gray matter. However, SUVmean and NSUVmean may be of potential interest for spinal cord diseases affecting predominantly the cord white matter as multiple sclerosis, Frieidreich's ataxia and VitB2 deficiency.
The longitudinal distribution of SUVmean an NSUVmean we observed closely matches that of SUVmax and NSUVmax. So far, the FDG uptake in PET/CT was measured in spinal cord tumors that are rare conditions in which a pattern of variably increased FDG uptake was reported [12]. FDG uptake in PET/CT of the spinal cord have also been applied to the evaluation of inflammatory myelopathies, including multiple sclerosis, neurosarcoidosis, and of amyotrophic lateral sclerosis [12,13]. While a pattern of decreased or increased FDG uptake was observed in inflammatory myelopathies and neurosarcoidosis, presumably reflecting varying delay between disease onset and time of PET and possible interference with steroid therapy [12,13], in case of amyotrophic lateral sclerosis elevated FDG uptake values were reported [19]. Additional potential diseases suitable to be evaluated with PET of the spinal cord include trauma, post-radiation myelopathy, Vitamin B12 deficiency myelopathy and some neurodegenerative diseases which primarily affect the spinal cord as Friedreich's ataxia and hereditary spastic paraplegia. We anticipate that in these conditions both SUVmax and SUVmean could provide valuable information because the cord white matter is frequently more affected than the central gray matter.
An issue addressed in the present study was the possible effect of normalizing the spinal cord SUV max and mean values to the liver FDG uptake. This type of normalization was originally proposed by Marini et al. [4] for spinal cord SUVmax in order to pool data provided by two PET/CT scanners. In our instance the normalization was justified by the need to overcome possible differences in FDG uptake between PET/CT and PET/MRI due to the time interval between the two acquisitions and the tracer decay. Remarkably, all the results we obtained by analyzing SUV max and SUVmean values in PET/MRI were confirmed when the data were normalized to the liver. This suggests that liver normalization might improve stability of results and might be useful for multi-center studies.
We recognize the following limitations of our study.
We evaluated the spinal cord FDG uptake in a population of adult subjects with lymphoma rather than of healthy subjects. However, performing PET/MRI in healthy subjects is unethical and the patients we enrolled had neither neurologic symptoms and signs nor MRI evidence of any spinal cord or marrow abnormality. Moreover, they were examined for initial staging purposes and had not received yet chemo or radiation therapies.
The sequential acquisition of PET/CT and PET/MRI could have determined lower spinal cord uptake values in the PET/MRI due to the tracer decay. Future PET/MRI studies of neurological conditions could avoid preliminary PET/CT acquisition that was requested by Health Authorities in Italy at the time of the present study.
We did not control for factors that in some prior studies were reported to potentially modify the FDG uptake as lean body mass, body surface area, serum glucose, insulin, free fatty acid, and exact time to acquisition. However, these factors are generally considered crucial in FDG repeat studies and not in cross-sectional studies like ours.
Finally, we used a whole-body protocol instead than a dedicated spinal MRI-PET. The latter could allow better morphologic definition of the spinal cord with improved co-registration and especially correlation with sequences for volumetric and microstructural assessment of the spinal cord. Of course, the combination of metabolic information from FDG PET with macro and microstructural information with high spatial resolution from MRI in the same examination will represent a valuable tool for improved understanding of the physiopathology of spinal cord diseases and might constitute a valuable surrogate marker for future trials. Also, exploration of other radiotracers than FDG may expand the role of spinal cord PET.

Conclusion
In conclusion, using a whole-body protocol we defined the maximum and mean FDG uptake of the normal spinal cord in PET/MRI. These values were higher than those measured in PET/CT so widening the dynamic range of this metabolic measurement. The SUV values showed the expected longitudinal distribution reflecting the segmental (metameric) functional specialization of the spinal cord. Normalizing the SUVmax and SUVmean of the spinal cord to the liver radiotracer uptake could be useful in multi-institutional comparisons and studies.

Ethics approval and consent to participate
This study was approved by the local ethics committee and informed consent was obtained from all the participants included in the study.

Consent for publication
Not applicable.

Availability of data and material
The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no conflict of interest.

Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Authors' contributions
Conception and design were performed by AM, SM and MM. Acquisition of data were performed by AV, CC and NE. Drafting of manuscript was provided SE, SA, PM, VA, DPR. Data analysis and interpretation were performed by AM, AV, CC, MM. All authors reviewed the manuscript. The authors read and approved the final manuscript.