18F-FDG PET/CT for the quantification of inflammation in large carotid artery plaques

Background There is currently no consensus on the methodology for quantification of 18F-FDG uptake in inflammation in atherosclerosis. In this study, we explore different methods for quantification of 18F-FDG uptake in carotid atherosclerotic plaques and correlate the uptake values to histological assessments of inflammation. Methods and Results Forty-four patients with atherosclerotic stenosis ≥70% of the internal carotid artery underwent 18F-FDG PET/CT. Maximum standardized uptake values (SUVmax) from all plaque-containing slices were collected. SUVmax for the single highest and the mean of multiple slices with and without blood background correction (by subtraction (cSUV) or by division (target-to-background ratio (TBR)) were calculated. Following endarterectomy 30 plaques were assessed histologically. The length of the plaques at CT was 6-32 mm. The 18F-FDG uptake in the plaques was 1.15-2.66 for uncorrected SUVs, 1.16-3.19 for TBRs, and 0.20-1.79 for cSUVs. There were significant correlations between the different uptake values (r = 0.57-0.99, P < 0.001). Methods with and without blood background correction showed similar, moderate correlations to the amount of inflammation assessed at histology (r = 0.44-0.59, P < 0.02). Conclusions In large stenotic carotid plaques, 18F-FDG uptake reflects the inflammatory status as assessed at histology. Increasing number of PET slices or background correction did not change the correlation. Electronic supplementary material The online version of this article (10.1007/s12350-017-1121-7) contains supplementary material, which is available to authorized users.


INTRODUCTION
Inflammation is a key factor in the pathophysiology of atherosclerosis with regard to progression and destabilization of plaques. 1 Patients with unstable carotid plaques have increased risk of plaque rupture and ischemic stroke, 2,3 and there is increasing interest in imaging carotid plaque inflammation in order to detect these unstable plaques.
Positron emission tomography (PET) imaging with 2-deoxy-2-( 18 F)-fluoro-D-glucose ( 18 F-FDG) of inflammation in atherosclerotic plaques has rapidly evolved since Rudd et al first reported 18 F-FDG accumulation in macrophage-rich areas of carotid artery plaques over a decade ago. 4 In contrast to oncology, 5 there is no consensus on methodological guidelines for 18 F-FDG PET/CT in atherosclerosis imaging. A recent position paper from the Cardiovascular Committee of the European Association of Nuclear Medicine 6 proposed optimized and standardized protocols for the imaging and interpretation of 18 F-FDG PET scans in atherosclerosis. However, they admitted that many of the recommendations suffer from the absence of conclusive evidence. Compared to a solid tumor, the cells responsible for 18 F-FDG uptake in carotid artery plaques are generally fewer, more dispersed, and spread around parts of the circumference of a tubular vascular structure. 3,7 Consequently, limited spatial resolution of the PET scanner and blood background activity are of great concern. Two parallel phenomena are known to influence measured activity in a lesion: 8 signal from the lesion lost to the surroundings (spill-out), and signal added to the lesion from the vessel lumen and adjacent anatomic structures (spill-in). 8,9 Different acquisition protocols and quantification methods have been suggested. [10][11][12][13][14][15][16][17] They all address the same concerns but have diverging solutions. A literature search identified 53 different acquisition protocols, 51 different reconstruction protocols, and 46 different quantification methods used in 49 studies. 9 The most common measure is the mean of all the maximum standardized uptake values (SUVs) (mean SUV max ) of the regions of interest (ROIs). The ROIs include the whole plaque (in localized stenosis), or one or more vessel segments (in subclinical/generalized disease). Bural et al calculated the atherosclerotic burden of the aorta by multiplying the mean uptake values for each aorta segment with the vessel wall volume. 18 Subanalysis looking for the most metabolically active areas of a vessel segment have been used in therapy response studies. 16 The uptake values are either normalized to the blood background activity, 15,17,19,20 corrected for the background activity with subtraction 11 or not corrected for background activity. 13,21,22 The rationale for background correction has been strongly criticized. 9,23 The aim of this study was to explore different methods for the quantification of 18 F-FDG uptake in large carotid artery plaques, and to correlate the uptake values to the amount of inflammation on histology.

Study Population
Forty-four patients referred to the Department of Neurology at our institution for the evaluation of carotid artery disease were included (Table 1). Inclusion criteria were ultrasound-confirmed atherosclerosis with internal carotid artery stenosis C 70% according to consensus criteria of the Society of Radiologists in Ultrasound. 24 Exclusion criteria were prior carotid endarterectomy/angioplasty with stenting, carotid occlusion, vasculitis, malignancy, prior radiation therapy to the neck, or immunotherapy. The Regional Committee for Medical and Health Research Ethics approved the study and all patients provided informed written consent.
Of the 44 included patients, 38 underwent carotid endarterectomy due to ipsilateral ischemic events or as prophylactic treatment before heart surgery. Eight of these plaques were lost to histological assessment due to logistical reasons.  The PET data were reconstructed to 2-mm slices with a pixel size of 2.67 mm using the ordered subset expectation-maximization 2D algorithm with four iterations (i), eight subsets (s)(4i/8s), and Gaussian post-reconstruction filter with fullwidth at half maximum of 3.5 mm. For attenuation correction, low-dose CT without the use of i.v. contrast was performed immediately before 18 F-FDG PET. For patients without a recent CT angiography of the carotid arteries, this was performed after 18 F-FDG PET.

F-FDG Uptake Quantification
The Hybrid Viewer 2.0 software (Hermes Medical Solutions AB, Stockholm, Sweden) was used for image fusion and 18 F-FDG uptake quantification. The CT angiography was used to guide drawing of the ROIs on the fused PET/CT slices. A plaque was defined as vessel wall thickening and a lumen contrast-filling defect on CT angiography ( Figure 1). 13 An experienced nuclear medicine physician (K.J.) drew ROIs around the entire vessel wall and lumen on all plaquecontaining axial PET slices ( Figure 2). ROIs were carefully placed to minimize the influence from 18 F-FDG uptake in structures close to the plaque (e.g., lymph nodes, paravertebral muscles, or salivary glands). Blood pool activity was obtained from four ROIs placed in the lumen of the jugular vein away from structures with 18 F-FDG uptake. Plaque localization in relation to the carotid bifurcation was recorded. The most cranial slice of the common carotid artery before the division was defined as the bifurcation. The pixel values in the PET images were converted into SUV normalized to lean body mass. 5 For all plaque ROIs, SUV max was obtained and the 18 F-FDG uptake was quantified using the following approaches ( Figure 2): (1) Max SUV max = the single highest SUV, (2) Mean SUV max = the mean of all plaque SUV max , (3) Most Diseased Segment (MDS)3 = the mean SUV max of the three contiguous slices centered on the slice with the highest SUV max , (4) MDS5 = the mean SUV max of the five contiguous slices centered on the slice with the highest SUV max , (5) Mean SUV max 4 = the mean SUV max of the four slices with highest SUV max .
For all blood pool ROIs, SUV mean was obtained. Blood background-corrected values for all the SUV measurements were calculated (target-to-background ratio (TBR); SUV divided by the blood pool activity (mean SUV mean in four venous regions) and corrected SUV (cSUV); blood poolcorrected SUV (subtraction of the blood pool activity (mean SUV mean ) from SUV).
A second independent experienced nuclear medicine physician (MER) drew ROIs on the 20 initial patients to assess inter-observer variability of the different quantification methods.

Endarterectomy and Histological Analysis
The mean time between PET/CT and endarterectomy was 6 days (range; 0-116, median; 0). The histological analysis has been described previously. 25 In brief, plaques were removed en bloc at carotid endarterectomy, fixed in 4% formaldehyde, decalcified in ethylenediaminetetraacetic acid, and cut into 2to 3-mm slices. After dehydration, the slices were embedded in paraffin and a 5-lm histological section from each slice was cut and stained with hematoxylin and eosin (H&E). The number of histology sections obtained from each plaque ranged from 2 to 9 (mean 4.9, SD 2.0). The plaques were assessed by a pathologist and a research physician blinded for the clinical and the 18 F-FDG PET findings. Inflammation was quantified using a modified version of the method used by Jander et al. 3 The sections were evaluated with 1209 magnification and the percentage area of inflammatory cells (macrophages and leucocytes) per plaque was obtained ( Figure 3): For each section, the total area and the area occupied by inflammatory cells were measured manually. The amount of inflammation per plaque was defined as the sum of all areas with inflammatory cells divided by the total area of all the sections. We performed repeated histopathological   assessment on selected sections from this study cohort and found that the amount of inflammatory cells was in the same 5% category at both assessments for 73% of the sections (Kappa = 0.73). 25 The amount of inflammation per plaque was 6.8% (SD 4.0; range 0.4-17.9).

Statistical Analysis
The SPSS Statistics software for Windows (IBM, version 21.0; SPSS Inc., Chicago, Ill) was used. Groups of data were compared using two-sided t-test and Pearson correlation for normally distributed variables. For non-normal distributions Mann-Whitney test and Spearman correlation were used. Statistical significance was set to 0.05.

Localization of Plaque and Highest 18 F-FDG Uptake
The length of the plaques in the cranio-caudal direction was 6-32 mm (mean 19, SD 7.6) and all included the carotid artery bifurcation (Figure 4). Max SUV max was located between 14 mm below to 18 mm above the bifurcation (mean 1.2 mm below the bifurcation) (Figure 4).

Inter-observer Variability
The correlation coefficients between 18 F-FDG uptakes calculated from plaques delineated by two nuclear medicine physicians independently were 0.96-0.98 for uncorrected SUVs, 0.63-0.68 for TBRs, and 0.90-0.93 for cSUVs. The correlation coefficient for background blood pool activity was 0.75.

Quantification of 18 F-FDG Uptake
Mean values, SDs, and ranges for the different quantification methods are summarized in Table 2. The 18 F-FDG uptake was significantly different for the different quantification approaches (paired samples ttest, P \ 0.004 for all pairs). TBR gave the highest mean values and the widest ranges, whereas the background-subtracted values (cSUV) showed the lowest mean values and the narrowest ranges.  Figure 4. Plaque extension (gray bars) and max SUV max location (black dots) per patient along the x-axis. The y-axis shows the distance in millimeter from the bifurcation (0) cranially in the internal carotid artery and caudally in the common carotid artery. 18 F-FDG uptake for the different quantification methods and corresponding background values for all patients are shown in Figure 5. Mean difference between max SUV max and mean SUV max was 0.08 and ranged from 0.02 to 0.66. The effect of background correction is shown in Figure 6. Correlations between 18 F-FDG uptake within the groups were 0.93-0.99 for the uncorrected SUV, 0.94-1.0 for TBR, and 0.92-0.99 for cSUV (Table 3).

Histology and 18 F-FDG Uptake
There were significant moderate (0.44-0.59) correlations between the amount of inflammation and all the different 18 F-FDG quantification methods (Table 4). The highest correlations were found for mean SUV max , and the lowest for max SUV max independent of background correction. Figure 7 shows scatter plots of inflammation versus max and mean SUV max with and without background correction.

DISCUSSION
In this clinical study, we explored different methods for the quantification of 18 F-FDG uptake in carotid plaques causing artery stenosis equal to or above 70%. 18 F-FDG uptake was homogenously disseminated throughout the entire plaques. Although there were differences in magnitude, quantification of 18 F-FDG uptake with and without background correction showed similar, moderate correlations to inflammation on histology.
Mean max SUV max (mean of the max SUV max for the study population) was only 13% higher than mean mean SUV max (Figure 8), whereas the three other quantification methods gave values in between, increasing with decreasing number of included slices. Homogenous 18 F-FDG uptake throughout the plaque contrasts findings from microPET of endarterectomized plaques showing patchy 18 F-FDG uptake. 26 The presence of macrophages reduces the thickness of the fibrous capsule and therefore an increasing number is likely to correlate to increasing vulnerability. As such, the highest 18 F-FDG uptake within a plaque could be the most appropriate parameter for risk assessment. Currently, the most used measure for assessment of plaque inflammation has been the mean SUV max . 17,21,22 No clinical studies have assessed the use of whole plaque max SUV max . Max SUV max is easily obtained, highly reproducible, and less influenced by partial-volume effects. 8,27 Although max SUV max is prone to image noise, 8 our findings suggest that for atherosclerotic plaque assessment in clinical PET, max SUV max should be explored further.
The strong correlation between the different quantification methods suggests that for this group of patients uncorrected 18 F-FDG uptake values may provide similar information as background-corrected values. As such, our findings do not support the superiority of TBR as quantification method as suggested by others. 4,10,[14][15][16][17] For circulation times above one hour, the blood background is low, but highly variable, whereas carotid plaques have consistent uptake over time. 11,13 A slight variation in blood background will therefore give significant variability in TBR. From a physics perspective, Huet et al 9 have explained that there is no legitimate rationale for using TBR instead of SUV because blood contamination is an additive and not a multiplicative process. A purely additive process would require subtraction of the mean luminal blood pool activity from the SUVs (cSUVs).
The inter-observer variability analysis of 18 F-FDG quantification revealed superior reproducibility of uncorrected SUV max compared to blood backgroundcorrected SUVs (TBR and cSUV). SUV measurements are highly dependent on the size, shape, and location of the drawn ROI because the ROI can either miss the voxel with the highest intensity or the ROI can inadvertently include contribution from adjacent 18 F-FDGavid organs. The use of contrast-enhanced CT to localize the plaques is likely to have contributed to the low interobserver variability in the SUV max in our study. Our correlation coefficient for blood background of 0.75 inevitably increases the inter-observer variability of all background-corrected values. Other 18 F-FDG uptake reproducibility studies in localized carotid artery plaques have reported moderate (background corrected) 28 to excellent (with and without background correction) 22,29 inter-observer agreement. The slightly inferior inter-observer agreement for background-corrected values found in our study could be related to differences in the placement of the ROI in the jugular vein as the diameter of the jugular vein often was small, making the measurement susceptible to image noise. However, this is an inherent limitation of all quantification methods with background correction.
The optimal method should predict plaque vulnerability and clinical outcome. In the present study, all quantification methods for 18 F-FDG uptake showed moderate correlation to inflammation on histopathology ( Table 4). The correlations were systematically slightly higher when mean SUV max was used instead of max SUV max . This was not unexpected as total plaque inflammation score and mean SUV max are both multislice methodologies. Total plaque inflammation score is a well-established method to correlate histology to ischemic symptoms, 2,3 or to 18 F-FDG uptake. 17 When comparing max SUV max with the slice with the highest percentage inflammatory area, the correlation did not increase (data not shown). This is in accordance with Tawakol et al 17 who found slightly higher correlations between overall plaque inflammation and histology than a slice-by-slice comparison of uptake value and histology. How to use histology as a gold standard to imaging is challenging. We know that excised plaques both shrink 17 and may be partly damaged. 17,25 The strength of our study was the close timing between the PET examinations and the endarterectomies. Our study patients had plaques not only presumed highly inflammatory giving recent symptoms, but also plaques removed prophylactically from asymptomatic patients.
A limitation of our study is the inclusion of patients with elevated blood glucose. Four had blood glucose values[ 11 mmol/L (198 mg/dL) and three 7-11 mmol/ L (126-198 mg/dL) at the time of the 18 F-FDG injection. Elevated blood glucose is known to reduce the uptake of 18 F-FDG into metabolic active cells in malignant diseases. 30 Guidelines for the clinical use of 18 F-FDG in inflammation and infection 31 also recommend the reduction of blood glucose to the lowest possible level. In studies on atherosclerosis, there is no consensus on a cutoff value. 12 Some studies do not report on blood glucose level, whereas in other studies the cut-off values have ranged from 8 mmol/L (144 mg/dL) 21 to 11.1 mmol/L (200 mg/dL). 16 Our correlations between histology and the different 18 F-FDG quantification methods were slightly increased when excluding the four patients with blood glucose level[ 11 mmol/L. We have not excluded patients with high blood glucose in the correlation analysis of the different uptake parameters and thereby we do not know if this has contaminated our results. In the 44 included patients, there was no correlation between blood glucose and background SUV mean . Another limitation is the wide range of circulation times (68-156 minutes) that could have influenced the plaque 18 F-FDG uptake by underestimating background-corrected values for patients with shorter circulation times and by overestimating background-corrected values for patients with longer circulation times. However, for 38 of the 44 patients the PET acquisition started between 85 and 115 minutes after 18 F-FDG injection, and thus, it is unlikely that difference in circulation time would change the findings in our study.

NEW KNOWLEDGE GAINED
Our study showed that SUVs without background correction from large plaques in the carotid artery can be used as inflammatory parameter in atherosclerosis.

CONCLUSION
In conclusion, in carotid artery stenosis equal to or above 70%, 18 F-FDG uptake reflects the inflammatory status as assessed on histology. Increasing number of PET slices or background correction did not improve the correlation.