Introduction

In the use of 2-deoxy-2-fluoro-D-glucose 18 F-FDG positron emission tomography–computed tomography (PET/CT) for cancer management, some form of quantification of tumour uptake is essential in order to monitor treatment. This is most often done by measuring the standardised uptake value (SUV), which is the fraction of administered activity per millilitre of tissue, multiplied by an index of body size, usually weight. An alternative approach is to scale tumour count to the count rate from the liver, which in this context is assumed to be relatively constant. Because tumour SUV is inversely related to blood glucose, European Association of Nuclear Medicine (EANM) guidelines recommend normalisation of tumour SUV to blood glucose of 5 mmol/l [1]. This approach assumes a tumour glucose utilisation rate (MRGLU) that is independent of blood glucose, in which MRGLU would be equal to the product of tumour FDG clearance (ignoring the lumped constant [LC]) and blood glucose. Thus, insofar as tumour SUV is a surrogate for tumour FDG clearance, the guidelines assume a hyperbolic relationship between tumour SUV and blood glucose. EANM guidelines also recommend rescheduling the study if blood glucose is above 7 mmol/l, although most departments would tolerate levels up to 10 mmol/l.

Liver SUV (SUVliver) is thought to be relatively constant [2, 3] and so an alternative approach is to express tumour count rate as a ratio with liver count rate [4–7]. The validity of this approach has received recent attention given the potential effect of hepatic steatosis on liver FDG uptake. Several groups have examined the relationship between steatosis and SUVliver. The results, however, were conflicting [8–12], likely because the effects of blood glucose on hepatic SUV were ignored.

FDG is dynamically exchanged between hepatocytes and blood (Fig. 1), resulting in a hepatic FDG concentration that closely mirrors the blood concentration [13–16]. Indeed, it has been recommended that in small animal work, the liver is a better region than the heart for monitoring the time course of blood activity [15]. Because blood FDG concentration is dependent on blood glucose [13], the hepatic FDG signal, rather than being constant, is also dependent on blood glucose, as has been previously shown [17, 18], such that the tumour-to-liver ratio also depends on blood glucose.

Fig. 1
figure 1

Model depicting the kinetics of FDG between blood and hepatocyte. FDG is phosphorylated in the hepatocyte (k3) but is not incorporated into glycogen. Following dephosphorylation (k4), it diffuses back into blood (k2). K1 represents hepatic blood flow

Full size image

The aim of this study was to compare the normalisation to blood glucose and scaling to the liver as correction procedures for quantification of tumour FDG uptake. Tumour SUV is highly variable in patients with cancer, so we used the brain as a surrogate for tumours to examine the effects of blood glucose. The justification for this relies on the premise that FDG uptake kinetics should be similar for tissues such as brain and tumour, but unlike liver, in which FDG is metabolically trapped.

In order to evaluate the impact of hepatic steatosis on scaling to the liver, we also compared the relationship between brain-to-liver SUV ratio and blood glucose in patients with and without hepatic steatosis. Finally, we explored an additional procedure that may be useful for correcting tumour FDG uptake for blood glucose.

Methods

Patients

This was a retrospective service evaluation based on 304 patients undergoing routine FDG PET/CT. We initially included a consecutive group of 156 patients over a two-year period (2011–2012). To increase the numbers of patients with low or high blood glucose levels, we included 148 additional patients with blood glucose levels > 6 (n =113) or < 4 (n = 35) mmol/l who underwent imaging between 2007 and 2012. Most patients were referred with cancer. Forty-seven patients were currently receiving chemotherapy and 56 had received chemotherapy previously.

All patients who visit our PET CT unit are asked to provide informed consent for use of any of their clinical data for publication, and all patients in this study did so.

PET/CT acquisition

Departmental protocol dictated that patients fast for at least 6 hours prior to their appointment. Blood glucose was measured using a glucometer (Accu-Chek Performa, Accu-Chek Inform II test strips; Roche Diagnostics, Indianapolis, IN, USA). FDG was injected intravenously via the antecubital fossa or hand. Patients were required to relax in a semi-recumbent position in a warm, quiet area during the uptake period. Imaging was performed 60 min post-injection of 400 MBq (±10 %) FDG.

A Siemens Biograph 64-slice PET with immediate non-enhanced CT imaging (120 kVp/50 mA—CARE Dose 4D; slice, 5 mm; pitch, 0.8; rotational speed, 0.5/sec), was used to cover the area from the orbital margin to the lesser trochanters. Arms were up, as arms down may result in artificial elevation of the liver FDG signal due to beam-hardening effects. Three-dimensional emission data were then acquired at 3 min per bed position (PET reconstruction: 4 iterations; subset, 8; Gaussian pre-filter; full width at half maximum (FWHM), 5 mm; matrix size, 168 × 168; zoom, 1).

Image analysis

Regions of interest (ROIs) were drawn over the right lobe of the liver (3 cm diameter, avoiding any focal lesions), frontal cortex (1.5 cm), basal ganglia (0.9 cm), and cerebellum (1.6 cm). The whole-body sweep did not include frontal cortex and basal ganglia in 107 patients. Maximum and average standardised uptake values (SUVmax and SUVave) were recorded for all four regions. An ROI identical to the liver ROI drawn on the PET image was drawn on the CT scan for measurement of CT density in Hounsfield units (HU). SUV in brain regions was multiplied by blood glucose and divided by 5 to give normalised SUV (nSUV).

Statistical analysis

The relationships between variables were quantified using Pearson’s linear regression analysis of continuous data. In addition, patients were categorised into six subgroups based on blood glucose within the following ranges (patient numbers in brackets): < 4 (59), 4–4.9 (76), 5–5.9 (38), 6–7.9 (38), 8–9.9 (58), and 10+ (35) mmol/l. Patients were also subdivided into those with hepatic steatosis (CT density < 40 HU; n = 71) and those without (>39 HU; n = 233) [19, 20].

Results

Relationship between blood glucose and hepatic CT density

There was an inverse relationship between blood glucose and hepatic CT density (R = −0.39, p < 0.0001; Fig. 2). Thus, patients with fatty liver tend to have high blood glucose.

Fig. 2
figure 2

Relationship between liver CT density (HU) and blood glucose. Straight line is linear fit

Full size image

Relationships of brain and hepatic SUV values with blood glucose

Cerebellar SUVmax declined non-linearly as a function of blood glucose (Fig. 3). The relationship was better fitted with an exponential function, the constant of which was −0.0993 mmol/l−1 (R = 0.75), than a hyperbolic function (R = 0.60). Corresponding exponential constants for frontal cortex and basal ganglia SUVmax were −0.119 mmol/l−1 (R = 0.76) and −0.114 mmol/l−1 (R = 0.75). Exponential constants for brain SUVave were similar: −0.113 (R = 0.77), −0.110 (R = 0.76), and −0.097 (R = 0.74) mmol/l−1 for the frontal cortex, basal ganglia, and cerebellum, respectively. In contrast, liver SUV increased non-linearly as blood glucose increased (Fig. 4).

Fig. 3
figure 3

Relationship between cerebellar SUVmax and blood glucose. Equations are shown for least squares fits to hyperbolic (y = K/x; bold line) and exponential (fine line) functions. Upper panel: linear ordinate; lower panel: logarithmic ordinate

Full size image
Fig. 4
figure 4

Relationships of brain SUVmax to liver SUVmax ratio with blood glucose. Vertical bars = SEM. Note that SEM values for the liver are too small to be visible

Full size image

Relationship between cerebellum-to-liver SUV ratio and blood glucose

The ratio of cerebellum-to-liver SUVmax declined non-linearly as blood glucose increased (Fig. 5). The relationship appeared hyperbolic (R = 0.83), as theoretically predicted (see Appendix; Eq. 12). The relationship between cerebellum SUVmax/liver SUVave and blood glucose was similar (data not shown).

Fig. 5
figure 5

Relationship between brain SUVmax (cerebellum) to liver SUVmax ratio (y) and blood glucose (x): y = 14.7/x

Full size image

Effect of hepatic steatosis on the relationship between cerebellum/liver SUV ratio and blood glucose

The relationships between cerebellum-to-liver SUV ratios and blood glucose were similar patients with and without hepatic steatosis, whether SUVmax or SUVave was the denominator (Fig. 6).

Fig. 6
figure 6

Relationships of brain SUVmax (cerebellum) to liver SUV ratios with blood glucose in patients with (closed circles, bold line) and without (open circles, faint line) hepatic steatosis. Upper panel: cerebellar SUVmax/liver SUVmax; lower panel: cerebellar SUVmax/liver SUVave. Least square fits to hyperbolic functions are shown. In both panels, the fits to steatosis and non-steatosis patients are similar

Full size image

Correction of brain SUV for blood glucose

Correction of tumour SUV for blood glucose by normalisation to 5 mmol/l (nSUV) aims to abolish the dependence of tumour SUV on blood glucose, and the same should apply to the brain. However, normalisation gave nSUVmax values for the cerebellum that correlated in a complex manner with blood glucose (Fig. 7a, upper panel). Thus, there was a positive correlation up to blood glucose level of 7 mmol/l, with a gradient of ~1 SUV per mmol/l (R = 0.44; p < 0.0001), but no further increase above 7 mmol/l. The overall relationship was therefore better fitted with a second-order polynomial (R = 0.55).

Fig. 7
figure 7

a Relationships with blood glucose of brain SUV normalised to blood glucose. Upper panel: cerebellar SUVmax normalised to blood glucose of 5 mmol/l. Straight line is linear fit to points up to blood glucose of 7 mmol/l. Bold curved line is second-order polynomial fit to all data. Lower panel: basal ganglia SUVave corrected for blood glucose by multiplication with e0.099 x glucose, where the exponential constant of 0.099 mmol/l was derived from the exponential relationship between SUVmax of the cerebellum and blood glucose (see Fig. 3). Correlation coefficients (R) are shown in each panel. b Relationships with blood glucose of brain SUV normalised to blood glucose (nSUV) and then divided by liver SUV. Upper panel: cerebellar nSUVmax divided by liver SUVmax; lower panel: cerebellar nSUVmax divided by liver SUVave. Patients with hepatic steatosis (HU < 40) are shown as open circles and those without (HU > 39) as closed circles. Straight lines are linear fits to all data points

Full size image

Division of cerebellar nSUVmax by liver SUVmax almost abolished its relationship with blood glucose, leaving a small gradient of 0.04 SUV per mmol/l (R = 0.17; p < 0.01; Fig. 7b, upper panel). There was no difference in this relationship between patients with and without hepatic steatosis. Results were similar when cerebellar nSUVmax was divided by liver SUVave instead of SUVmax, although the correlation coefficient and regression gradient were slightly higher (Fig. 7b, lower panel).

Based on the exponential constant of −0.099 mmol/l−1 derived from the relationship between cerebellum SUVmax and blood glucose, multiplication of basal ganglia SUVave with e0.099.glucose gave corrected SUV values that showed no correlation with blood glucose (R = 0.03; Fig. 7a, lower panel). Application of this correction procedure to the frontal cortex yielded similar results (data not shown).

Discussion

Brain SUV as a function of blood glucose

FDG enters the cell using the same transport system and phosphorylation enzymes as glucose, but is then metabolically trapped. As Hasselbalch et al. [21] found no change in regional cerebral MRGLU in response to acute hyperglycaemia, elevated blood glucose levels should result in decreased FDG uptake in the brain. Accordingly, we found an inverse nonlinear relationship between brain SUV and blood glucose that was well fitted by an exponential function. An inverse relationship between brain SUV and blood glucose has been shown previously in works by others [22, 23]. Buchert et al. [24] showed that acute reversal of hyperglycaemia increased FDG uptake in the brain by up to 80 %, while Kawasaki et al. [25] found that mild hyperglycaemia affected the distribution of FDG uptake in the brain of normal subjects. Critically, exponential constants of the relationships between brain SUV and blood glucose in the current study were similar for the cerebellum, basal ganglia, and frontal cortex.

Validity of liver SUV for scaling tumour SUV: effects of steatosis and blood glucose

Liver SUV, in contrast to brain SUV, showed a positive relationship with blood glucose (Fig. 4). The brain/liver SUV ratio, therefore, showed a strong inverse relationship with blood glucose (Fig. 5). As theoretically predicted (Appendix), this was hyperbolic. Although much attention has recently been focussed on the influence of hepatic steatosis on the validity of liver SUV as a comparator for tumour FDG uptake [8–12], blood glucose clearly has a more dramatic effect than hepatic fat on the brain-to-liver SUV ratio. Recent studies addressing the validity of the fatty liver as a comparator did not consider the influence of blood glucose on hepatic SUV, even though patients with fatty liver disease tend to have high blood glucose, as shown in Fig. 2. Two recent studies from our group [26, 27] confirmed that when blood glucose is taken into account, FDG accumulation in the liver is increased in hepatic steatosis. However, as shown in the current study, steatosis has much less impact on scaling brain SUV to liver SUV in comparison to blood glucose.

Use of brain SUV as a surrogate for tumour FDG uptake

Hasselbalch et al. [21] measured MRGLU in the brains of normal subjects using the Sokoloff model. The authors showed that during acute hyperglycaemia in comparison to normoglycaemia, there was no change in global or regional MRGLU in cortical and subcortical grey matter. Most studies [23, 28–30], though not all [31, 32], have shown that tumour SUV also decreases at high blood glucose levels, as is recognised in both the EANM and the Society of Nuclear Medicine guidelines [1, 33]. Lindholm et al. [28] showed no change in tumour MRGLU in response to oral glucose loading in five patients with head and neck cancer. Moreover, Crippa et al. [29] recorded a decrease in mean SUV of 20 hepatic colorectal metastatic deposits from 9.4 to 4.3 in a study in which patients underwent imaging on two separate occasions: fasting, and following a glucose load that increased blood glucose from 5 to 9 mmol/l. Ishizu et al. [23] recorded a decrease in brain glioma SUV from 4.41 to 1.54 associated with an increase in blood glucose from 5.9 to 13.5 mmol/l. Insofar as the brain can be regarded as a surrogate for tumours, in that both display no change in MRGLU in relation to glucose levels, the liver clearly cannot be used as a tissue comparator for tumour FDG uptake.

Correction of tumour SUV for blood glucose

The similarity between brain and tumour FDG kinetics can be extended to the correction of tumour SUV for blood glucose. EANM guidelines recommend normalising tumour SUV to 5 mmol/l [1]. To be successful, this procedure is required to abolish any relationship between tumour SUV and blood glucose. Its validity in this respect is supported by some studies [34, 35] but not by others [31].

A critical assumption underlying the use of the brain as a surrogate for tumour FDG uptake is that MRGLU in the brain and tumours is independent of blood glucose. There is little robust quantitative data in the literature concerning the effects of blood glucose on tumour SUV. Assuming an inverse exponential relationship between tumour SUV and blood glucose, it can be calculated from the data of Lindholm et al. [28] that the exponential constant for tumours would be −0.1 to − 0.15 mmol/l−1. Based on the data of Crippa et al. [29], the constant would be about − 0.2 mmol/l−1, although their SUV values had very high standard deviations. Based on the data of Ishizu et al. [23], the constant would be −0.093 mmol/l. These estimated exponential constants are broadly similar to those recorded here for the brain (−0.099 to −0.119 mmol/l−1).

Given the variation in tumour SUV, it is difficult to determine the relationship between blood glucose and tumour SUV in large heterogeneous patient populations. Nevertheless, in such an attempt, Zhuang et al. [30] recorded a weak inverse correlation between tumour SUV and blood glucose, without indicating whether the relationship was hyperbolic, exponential, or linear. In contrast, in 248 patients with lung cancer, Hallet et al. [31] found no significant correlation between tumour SUV, expressed logarithmically (thereby implying an exponential relation), and blood glucose. Differing levels of metabolic activity across tumours would not rule out a unique exponential constant that could be universally applied to tumours, provided that tumour MRGLU, like brain MRGLU, is independent of blood glucose.

In the current study, normalisation failed to achieve independence of the brain nSUV in relation to blood glucose, and instead generated a significant nonlinear correlation (Fig. 7). However, when brain nSUVmax was divided by liver SUVmax, it showed almost no relationship with blood glucose, regardless of whether patients had hepatic steatosis (Fig. 7). This finding is consistent with Eq. 12 (Appendix), which shows that (tissue SUV [SUVi] × blood glucose)/SUVliver is constant. When brain nSUVmax was divided by liver SUVave instead of SUVmax, a moderate linear positive correlation remained with blood glucose. We believe this is the result of a ‘signal dilution’ effect on liver SUVave of excess hepatic fat, the distribution of which is not homogeneous [36]. Such a dilutive effect is less with SUVmax than SUVave because SUVmax is selectively based on a voxel that is relatively free of fat [26]. The correlation between brain nSUVmax/liver SUVave and glucose is then the result of higher levels of hepatic fat in patients with high blood glucose. As such, for this dual correction procedure, liver SUVmax is recommended. The sense of the procedure is evident when, for example, there is heavy uptake of FDG by bone marrow. This would reduce the availability of FDG uptake elsewhere and reduce left ventricular blood SUV (SUVLV) and SUVliver. Division of nSUV by SUVliver would then appropriately increase nSUV. Another attractive feature of this approach is that it avoids the emerging issue of the best whole-body index—weight, lean body mass, or body surface area—upon which to base SUV [37], since whatever index used for measuring SUV is cancelled out.

Multiplication of SUV by e0.099 x glucose (based on the exponential constant of 0.099 mmol/l−1 for the relationship between cerebellar SUVmax and blood glucose) is another potential correction procedure. It abolishes the relationship between basal ganglia SUV and blood glucose (Fig. 7a). If a constant of approximately −0.1 mmol/l−1 is not universal for tumours, this correction would then not be valid for tumours. Further work is needed to address this issue, particularly in regard to the independence of tumour MRGLU and blood glucose level.

Study limitations

Limitations of the study include its retrospective design and heterogeneous patient population. Whilst every effort was made to avoid any focal liver pathology when drawing the liver ROI, it is possible that covert pathology, such as diffuse colorectal micrometastases, may have increased liver SUV. Moreover, the frontal cortex and basal ganglia were included in the whole-body sweep in only two-thirds of the patients, although patient numbers were still high.

Conclusions

In conclusion, tumour FDG uptake cannot be factored to liver SUV because of the effects of blood glucose rather than any influence of hepatic fat. Normalisation of brain SUV to blood glucose results in a complex relationship with blood glucose and is therefore also invalid. Combining the two correction procedures, however, gives corrected values that show almost no relationship with blood glucose, and the success of this dual correction procedure is explained theoretically in the Appendix. Normalisation to blood glucose using an exponential constant could also be useful.