Correlations Between 18F-FDG PET/CT Parameters and Pathological Findings in Patients With Rectal Cancer.

Correlations Between

18

F-FDG PET/CT

Parameters and Pathological Findings in

Patients

With Rectal Cancer

Chih-Ying Liao, MD,* Shang-Wen Chen, MD,Þþ§ Yi-Chen Wu, MD,?

William Tzu-Liang Chen, MD,||

Kuo-Yang Yen, BS,**ÞÞ Te-Chun Hsieh, MD,**ÞÞ Pin-Jie Chen, MD,þþ

and Chia-Hung Kao, MDþ**

C

mainly on the clinical stage or tumor location.2 For patients whose

clinical diagnoses fall under T3YT4 or a positive node as defined by the American Joint Committee on Cancer (AJCC), the treatment of choice is always preoperative chemoradiotherapy (CCRT) followed by total mesorectal excision.3 By contrast, patients at an earlier stage

of the disease may receive straightforward curative surgery, whereas adjuvant treatment is reserved for patients displaying certain highrisk pathological features.4

Currently, the modalities used most often to identify candidates for preoperative CCRT are CT, endorectal ultrasound (EUS),

and MRI. Of these, EUS and MRI have been shown to provide superior detection of tumor invasion or regional node metastasis.5Y7

18F-FDG PET or PET/CT has become increasingly popular for tumor

staging and to assess a patient’s response to preoperative CCRT. Two studies have reported that an assessment of FDG PET findings altered the treatment scheme for approximately 8% to 17% of patients.8,9 On

the other hand, the associations between PET data and pathological findings have not been thoroughly investigated. Although 3 studies have correlated PET-related parameters with pathological findings for rectal cancer,10Y12 no study has compared comprehensive

predicting invasion depth or the final pathological stage. A unique advantage of FDG PET is its ability to automatically

create a tumor contour using quantitative information within the tumor. In addition to the application of SUVmax, the predictive roles

of metabolic tumor volume (MTV) and total lesion glycolysis (TLG) have been investigated as predictors of treatment outcome.13,14

Compared with CT- or MRI-based tumor delineations, which are mainly based on anatomical data, the autocontouring process of PET not only reduces interobserver variability in image evaluation but also provides detailed biological information.15 For this study, we

examined the correlations between various autosegmentations and pathological tumor size and stage to determine the most suitable method. Our results add to the existing knowledge on the use of EUS or MRI to identify patients for preoperative CCRT or aggressive adjuvant therapy.

MATERIALS AND METHODS

Patients

The patient group included 67 patients with newly diagnosed

cancer of the rectum or rectosigmoid junction. All patients were seen at China Medical University Hospital between January 2009 and December 2010, and the study design was retrospective. The study was approved by the local institutional review board (certificate no. DMR99-IRB-010-1). All patients were scheduled to undergo curative surgery, with preoperative PET/CT as an allocated workup for cancer staging. According to the treatment protocol of our hospital, all patients with rectal cancer diagnosed as AJCC clinical stage T3YT4 or with positive lymph nodes evident on the CT or MRI are strongly advised to receive initial preoperative CCRT. However, patients may decline this protocol and choose curative surgery as the

initial treatment. Thus, no patients in the study cohort received preoperative CCRT The median age of the patient group was 58 years (range,

30Y86 years), and 39 patients were men and 28 were women. Tumor locations were classified as rectum in 44 patients, and at the rectosigmoid junction in 23 patients. Surgical resection was performed within

28 days following PET/CT. Because diabetes patients were excluded from the initial PET/CTworkup, all patients in our group had a reference serum glucose level before PET/CT imaging. The characteristics of the 67 patients are shown in Table 1.

PET/CT Image Acquisition

All patients fasted for at least 4 h before 18F-FDG PET/CT

imaging. The images were captured using a PET/CT scanner (PET/ CT-16 slice, Discovery STE; GE Medical Systems, Milwaukee, Wis) approximately 60 minutes after the administration of 370 MBq of

18F-FDG. Patients were requested to rest during the uptake period.

The PET/CT workstation provided a quantification of FDG uptake for SUV. A nuclear medicine physician identified the locations and the SUVmax value for all primary tumors. A second physician confirmed the point of SUVmax within the tumors. This procedure has

been described in our previous reports.14,16

Measurement of Maximum TL and Width in PET/CT

The PET/CT-based maximum tumor length (TL) and tumor

width (TW) were measured on attenuation-corrected FDG PET images. The measurements were calculated using an SUV-based automated contouring program (Advantage Workstation Volume Share

version 2; GE Health). The FDG PET data in DICOM format were

inputted into the workstation, and the images were reviewed to localize the target lesions, with the site confirmed by 2 nuclear medicine physicians. The physicians were blinded to data gained from the preoperative image and pathological findings. To define the maximum TL and TW surrounding the tumor, we used the measured

distance greater than 20% of SUVmax (TL20% and TW20%), 30% of SUVmax (TL30% and TW30%), 40% of SUVmax (TL40% and TW40%), and 50% of SUVmax (TL50% and TW50%). The unit of calculation was millimeters. Figure 1 shows the contours of various thresholds of TL and TW for a patient with rectal cancer.

Measurement of MTV and TLG

The MTVs were measured from attenuation-corrected FDG

PET images using an SUV-based automated contouring program (Advantage Workstation Volume Share version 2; GE Health). The FDG

PET data in DICOM format were inputted into the workstation, and these images were reviewed to localize the target lesions, as confirmed by 2 nuclear medicine physicians. Metabolic tumor volume was defined as the sum of metabolic volumes of the primary tumors. The

volume boundaries were drawn sufficiently wide to incorporate each target lesion in the axial, coronal, and sagittal FDG PET images. To define the contouring margins surrounding the tumor, we

used SUVmax of 2.5 (MTV2.5) and SUVmax of 3.0 (MTV3.0), as previously reported.14 Furthermore, volumes greater than 20% of

SUVmax (MTV20%), 30%of SUVmax (MTV30%), 40% of SUVmax (MTV40%), and 50% of SUVmax (MTV50%) were also measured. The contour surrounding the target lesions within the boundaries was automatically produced, and the voxels presenting an SUV intensity of SUVmax Q 2.5 or Q 3.0, and Q 20% Q 30% Q 40% or Q 50% of

SUVmax within the contouring margin were incorporated to define the tumor volumes. The MTVs for the primary tumors included adjacent lymph nodes with a small volume. The small lymph nodes adjacent to the primary tumor cannot be segmented from the primary tumor on PET/CT if they appear similar to the part of the primary lesion. However, large lymph nodes neighboring the primary tumor can be segmented using an automatic VOI tool on PET/CT, even if the nodes are

partially contiguous to the primary tumor. Although visual interpretation can be used to evaluate MTV,17 our study excluded this method

because it is susceptible to variations caused by window-level settings and is highly operator-dependent.18 These influences may result in a

wide variability among institutions.

The TLGs were also measured from attenuation-corrected

FDG PET images using an SUV-based automated contouring program (Advantage Workstation Volume Share version 2; GE Health),

as described. The TLG was calculated according to the following formula: TLG = mean SUV _ MTV. We used threshold levels that were equivalent for the MTVs, that is, TLG2.5, TLG20%, TLG30%, TLG40%, and TLG50%.

Pathology

A routine pathology preparation was performed following tumor

resection. The examination was conducted by a pathologist who specialized in gastrointestinal cancer. The maximum length and width of

the dissected primary tumors (abbreviated as pTL and pTW, respectively) were carefully measured macroscopically using a ruler before

slicing. The measurements were confirmed by a second pathologist. The pathologists were blinded to any information from PET/CT tumor measurements. Figure 2 shows the measurement of pTL and pTWin a specimen from a patient with rectal cancer.

Statistical Analysis

SD, unless otherwise indicated. Tumor diameters were measured as TL and TW at various thresholds, and Pearson correlation coefficients (r) were calculated between these values and the pTL and pTW values. To examine the correlations between the parameters and AJCC pathological T and N stages, receiver operating characteristic curves were created to evaluate the optimal predictive performance among the TWs, MTVs, and TLGs. The predictive values of PET/CT-related

parameters for the pathological stage were further examined using logistic

regression analysis. All analyses were 2-sided, and P G 0.05 was considered statistically significant. Statistical analyses were performed

using SPSS, version 13.0 (SPSS Inc, Chicago, Ill).

RESULTS

Correlation Between pTL/pTW and PET/CT

Parameters

The pTL values for all patients ranged from 2.0 to 8.0 cm for pathology examination (4.4 T 1.3 cm). The pTW values ranged from 1.3 to 6.5 cm (3.4 T 1.0 cm). The results of correlational analysis for various thresholds of PET/CT-based TLs and TWs are shown in Tables 2 and 3, respectively. The parameters TL30% and TW40%, respectively, showed the best match with pTL (r = 0.72, P G 0.001) and pTW (r = 0.44, P G 0.001). The distribution of TL30% against pTL is shown in Figure 3, and that of TW40% against pTW is shown in Figure 4. The relationship between TL30% and pTL showed a considerably stronger correlation (more linear) than that between TW40% and pTW.

Correlation Between Pathological T Stage and PET/

CT Parameters

Patients were divided into 2 groups according to their pathological

T stage: T1YT2 (pT1YpT2, n = 14) and T3YT4 (pT3YT4, n = 53). The receiver operating characteristic curves were analyzed

to compare the efficacy of various methods for determining thresholds for autosegmentation contouring, and the results showed that

MTV2.5, MTV40%, TLG40%, and TW40% predicted the pT3YT4 stage most accurately. These parameters were compared with each other, and the areas under the curve were 0.88 T 0.06, 0.83 T 0.06, 0.86 T 0.07, and 0.68 T 0.09, respectively (Supplemental Figure, http://links.lww.com/CNM/A2). Using the median value of MTV2.5 (28 mL) as a cutoff, the sensitivity and specificity for predicting

pT3YT4 were 64.1% and 92.8%, respectively (positive predictive value = 97.1%, negative predictive value = 40.6%, accuracy =

70.1%). These results showed that MTV2.5 provided the best specificity and positive predictive value for pT3YT4.

As displayed in summary form in Table 4, logistic regression analysis showed that MTV2.5 was the only significant predictor of pT3YT4 (P = 0.001; odds ratio [OR], 1.81; 95% confidence interval [CI], 1.26Y2.60). The MTV2.5 values for the pT1YpT2 and pT3YT4 patient groups were 11.6 T 11.4 and 34.6 T 21.4 mL, respectively. The distribution of MTV2.5 stratified by the pathological T stage is shown in Figure 5. The TLG40% values showed a marginal effect in predicting pT3YT4, with an OR of 1.08 (P = 0.08; 95% CI, 0.98Y1.15).

Patients were also dichotomized according to their pathological N stage: node-negative (pN0, n = 37) and node-positive (pN1Y3, n = 30). However, none of the PET/CT parameters showed a significant difference between the 2 groups.

DISCUSSION

Interest in the role of FDG PET in tumor staging5,7 or monitoring

the response to treatment19,20 in patients with rectal cancer is

growing. However, optimal methods derived from PET or PET/CT parameters in tumor mapping or assessing local invasion have not been widely investigated. This was the first study to examine the feasibility of various autosegmentation methods based on correlations with pathological references. We compared various threshold

methods for autosegmentation to simplify image acquisition and minimize interobserver differences. Our results showed that a crude estimation of pathological diameters was achieved by simply measuring the parameters from preoperative PET/CT, using the TL30%

and TW40% methods. Higher MTV2.5 values showed the strongest correlation with pT3-T4 staging, which could help identify patients who would benefit from preoperative CCRT or more aggressive treatment. This finding has not been reported before and will assist in optimizing treatment schemes.

To evaluate the extent of local invasion in patients with rectal cancer, MRI reportedly displays excellent sensitivity in detecting transmural penetration using the phase-arrayed coil.5,21 MRI also

general, MRI is considered a promising tool in assessing locally advanced diseases.5 By contrast, EUS is a sensitive tool for early rectal

cancer detection, with an accuracy of 69% to 97%.5,7 Only 2 studies

have investigated the role of PET in tumor mapping, possibly because of the limited spatial resolution of this method.

Ciernik et al10 investigated a series of 11 patients with rectal

cancer whowere undergoing preoperative radiation therapy. The results

showed that the PET-related volume with a single-peak threshold of 40% was strongly correlated with the CT-derived tumor volume. Similarly,

Buijsen et al11 conducted a comparative analysis using a cohort

of 26 patients who received PET/CT, CT, and MR. Their study used an automatic contouring method according to source-to-background ratios to find an optimal percentage threshold within the defined volume for each patient. The results showed that CT scans tended to overestimate the TL. The correlations between imaging and pathological parameters were stronger forMRI than CT, but automatic PET measurements were most strongly correlated with pathology. One study assessed the accuracy of PET/CT in staging by comparing the changes in SUVmax and

showed that post-CCRT PET/CTwas an inadequate tool for evaluating anatomic tumor changes and could not accurately predict tumor clearance of the mesorectal fascia.22

Two studies have shown that commonly used imaging modalities, including FDG PET, MR, EUS, and CT, are limited in malignant node identification.5,23 An earlier study showed that PET displayed

particularly poor sensitivity for nodal diseases in cases of CRC.22 This

finding was consistent with ours, in that our results showed that the PET/CT parameters of primary tumors did not distinguish the presence of pathological nodes. This finding might be attributable to the low spatial resolution of PET, or that small lymph nodes adjacent to primary lesions cannot be segmented separately from the primary tumor when they appear similar to a part of the primary lesion.

This study was subject to certain limitations. First, the maximum macroscopic tumor diameters were measured after formalin fixation rather than being mapped immediately after tumor resection. Tumor shrinkage thus might have biased the correlations to an unknown extent. As described by Goldstein et al,24 CRC surgical specimens may

shrink by up to 30% after fixation of the specimen. Jonmarke et al reported linear correction factors for tumor shrinkage between 1.04 and

1.14 in cases of prostate cancer.25 The results of our correlational study

would have been more robust if we could have reported a correction factor for tumor shrinkage. Second, because the rectum is a hollow

organ, residual stool or air in the lumen might have affected the measurement of PET-related parameters. Third, the position between 2

measurements might be not entirely comparable because the pathologists were blinded to information from PET/CT tumor measurements.

Preferably, the resected specimen should be oriented to the same position when the scan was done, which would be difficult for tumors located at the upper third or rectosigmoid junction. These 3 limitations were probably the main factors detracting from the linearity of the correlations between TL30% and TW40% and the pathological diameters. Finally, our study investigated an optimal PET/CT approach

for matching images and pathological variables. Thus, we did not compare PET/CT data directly against those of other radiological images.

The advantage of PET/CT over other imaging modalities should be further investigated. In addition, further studies are essential to minimize the confounding influence of partial volume effects on PET,26

particularly for tumors with smaller volumes.

Despite these limitations, our study represented a novel approach in forecasting pathological findings from preoperative PET/

CT. Because the optimal thresholds are always determined by specific PET centers, type of malignancy, and tumor characteristics,

future studies should include more patients and should use a prospective study design. Also, long-term patient outcome using these

parameters should be investigated to optimize their predictive value. Furthermore, innovative methods should be assessed, such as accurate 3-dimensional volume measurements,27 image comparison, and

bowel preparation.

CONCLUSIONS

For patients with rectal cancer requiring surgical procedures, preoperative PET/CT can be used as a supplemental tool to predict pathological findings. The parameters TL30% and TW40% were correlated with the maximum diameters of the surgical specimens. In addition, a higher MTV2.5 value was associated with high specificity for an advanced T stage, for which preoperative CCRT should be considered initially