Predictive Value of [18F]FDG PET/CT for Lymph Node Metastasis in Rectal Cancer

[18F]Fluorodeoxyglucose ([18F]FDG) Positron emission tomography/computed tomography (PET/CT) is commonly used for rectal cancer staging, but improved diagnostic methods for nodal metastases are needed. We aimed to evaluate whether the combination model of the metabolic tumor volume of primary tumor (T_MTV) and maximum standardized uptake value of lymph node (N_SUVmax) on pretreatment [18F]FDG PET/CT could improve nodal metastases prediction in rectal cancer. We enrolled a total of 166 rectal cancer patients who underwent pretreatment [18F]FDG PET/CT and surgical resection without neoadjuvant treatment between January 2009 and August 2016. Visual and semiquantitative PET/CT parameters were obtained. Associations between clinicopathological, PET/CT-derived variables and nodal metastases were evaluated by logistic regression analysis. Nodal metastases were confirmed histologically in 68 of the 166 patients (41%). Uni- and multivariate analyses demonstrated T_MTV and N_SUVmax were independent predictive factors for nodal metastases. The c-statistics of the combination model was 0.806 (Standard Error, 0.034; 95% Confidence Interval, 0.737–0.863), which showed significant improvement compared to T_MTV (0.698, P = 0.0002) or N_SUVmax (0.720, P = 0.0008) alone. T_MTV and N_SUVmax are independently correlated with nodal metastases. Furthermore, the combination model showed improved performance for risk prediction; thus, [18F]FDG PET/CT might have a role in rectal cancer staging and treatment planning.


Results
patient Characteristics. A total of 166 patients with rectal cancer who received curative surgical resection were retrospectively analyzed. The characteristics of the enrolled patients (mean age, 66.7 ± 10.6 years) and the associations with LN metastases are listed in Table 1. LN metastases were confirmed histologically in 68 patients (41%), and 98 patients (59%) presented with no LN metastases. Although the majority of patients were diagnosed with 7th American Joint Committee on Cancer (AJCC) stage I-III cancer, two patients with pathologic stage Tis were classified as stage 0, and one patient with distant metastasis was classified as stage IV. The stage IV patient had 6 nodal metastases (N2a) and underwent a low anterior resection of a single hepatic metastasis and left hepatectomy. Pathologic stages according to the 7th AJCC and PET parameters, such as SUVmax of primary tumor (T_SUVmax), T_MTV, and N_SUVmax, were significantly different between the two groups; however, no significant differences were found with respect to pre-operative carcinoembryonic antigen (CEA), pathologic tumor size, and histologic grade. Thirty-nine of the 166 patients (23.5%) showed positive nodal uptake by [

Uni-and Multivariate Analyses.
Univariate logistic regression analysis revealed that T_SUVmax, T_MTV, and N_SUVmax were significantly associated with LN metastases ( Table 2). In the multivariate analysis, both T_MTV (odds ratio [OR], 1.022; 95% confidence interval [CI], 1.001-1.043; P = 0.038) and N_SUVmax (OR, 2.181; 95% CI, 1.523-3.125; P < 0.001) were found to be significant predictive factors; otherwise, T_SUVmax was justifiably removed from the stepwise model. These two independent parameters were used to construct a nomogram for risk prediction of LN metastasis (Fig. 1). To use the nomogram, the points for each parameter should be determined by drawing a vertical line from the exact value of the variables to the points row. Then, total points are calculated by arithmetic sum. Finally, the individual predictive probability of LN metastasis can be obtained by drawing a vertical line from the total points row to the probability of LN metastasis row.

Discussion
In the current study, we assessed the diagnostic value of metabolic parameters measured by [ 18 F]FDG PET/CT for the prediction of LN metastases in patients with rectal cancer. Our results demonstrated that nodal [ 18 F]FDG uptake findings were highly specific for LN metastases status, but it had a limitation due to its relatively low sensitivity. To overcome this limitation, we used metabolic parameters such as T_MTV and N_SUVmax for precise diagnosis of LN metastases. T_MTV and N_SUVmax are independent predictive factors for LN metastases in patients with rectal cancer. Moreover, the combination of both parameters significantly improved LN metastases prediction beyond each independent parameter alone (Fig. 3).
The sensitivity of [ 18 F]FDG PET/CT was found to be relatively low (48.5%), although the specificity was high (93.9%). This finding is similar to that in previous studies, which showed poor sensitivity for detecting LN metastases 12,19 . In this study, we excluded the patients who had received neoadjuvant treatment, since any treatment before surgical resection could affect the histopathologic results, including the initial LN status. If these advanced rectal cancer patients who underwent neoadjuvant chemotherapy were included in the present study, the LN detectability of [ 18 16 . However, they only focused on metabolic activity of the primary tumor and did not use visual or semiquantitative information from [ 18 F]FDG-avid LNs as a possible predictive factor for LN metastasis. In contrast, we evaluated not only metabolic information from the primary tumor, but also the LN's own metabolic activity, for LN metastases prediction. For semiquantitative analysis of [ 18 F]FDG-avid LNs, we adopted N_SUVmax which is the most widely accepted parameter of [ 18 F]FDG PET/CT. Because SUVmax is a simple measurable metabolic parameter, and the application of metabolic volume parameter has a limitation in small volumes 20 . Tsunoda et al. demonstrated that N_SUVmax could improve the accuracy of preoperative LN metastases detection when compared to nodal diameter 12 . Concordantly, N_SUVmax was an independent prognostic factor for LN metastases in this study. We, therefore, have incorporated N_SUVmax as well as T_MTV for LN metastases prediction. Additionally, the combination of T_MTV and N_SUVmax could improve LN metastases prediction in patients with rectal cancer.
The present study also suggested that T_MTV could be a useful complementary predictive factor itself for LN metastases. Recently, several studies have shown that volumetric parameters measured by [ 18 F]FDG PET/CT are useful for the evaluation of therapeutic response or prognostication in a variety of malignancies [21][22][23][24][25] ; T_MTV can be used as a predictive factor for LN metastases in lung, endometrial, and uterine cervical cancers 18,26,27 . Our result is consistent with previous studies in that T_MTV obtained by [ 18 F]FDG PET/CT could be an effective marker of total tumor burden and may reflect the aggressiveness of cancer associated with LN metastases. A previous meta-analysis identified numerous histopathological factors that may be correlated with LN metastases in primary CRC 28 . However, no single histopathological feature reliably predicted LN metastases, and these factors can be evaluated only after surgery. Considering the feasibility of [ 18 F]FDG PET/CT in the pre-operative setting This study had a few limitations. First, the single-center retrospective design of this study might be subject to selection bias. Further studies are needed to validate the results of the present study. Second, physiologic [ 18 F] FDG uptake in the gastrointestinal tract may cause overestimation of T_MTV. Therefore, we adopted an absolute SUV threshold of 2.5 for MTV measurement, which was a widely used cutoff value in several previous studies and could reduce inter-and intra-observer variation in delineation of tumor volume using a software-assisted automatic method 21,32 . Lastly, we did not correct for a partial volume effect, which may have underestimated the value of SUVmax; this is because partial volume correction is generally too complex to use in daily clinical practice. New and feasible partial volume correction methods would be helpful in achieving precise quantification and clinical application 33,34 . Despite these limitations, we have shown that the combination of T_MTV and N_SUVmax could be an attractive strategy to further improve the diagnostic performance of [ 18 F]FDG PET/CT for LN metastases in patients with rectal cancer.
In conclusion, T_MTV and N_SUVmax were independent prognostic factors for the prediction LN metastases in rectal cancer patients. Furthermore, our prediction model using T_MTV and N_SUVmax could provide a more precise prediction of LN metastases. The use of N_SUVmax in combination with T_MTV on preoperative [ 18 F]FDG PET/CT could be a useful tool for initial staging and treatment planning in patients with rectal cancer.

patients.
Between January 2009 and August 2016, the medical records of 296 consecutive patients who underwent surgery for rectal cancer and had a preoperative [ 18 F]FDG PET/CT were evaluated retrospectively. Of these, patients who received neoadjuvant chemoradiotherapy (n = 96), endoscopic tumor removal prior to surgery (n = 24), or long delayed interval over than 1 month between [ 18 F]FDG PET/CT and surgery (n = 10) were excluded. A total of 166 patients were enrolled in this study. Surgery (160 patients with anterior resection, 4 with abdominoperineal resection, 1 with Hartmann's operation, and 1 with total proctocolectomy) was performed by experienced colorectal surgeons, which included total mesorectal excision and at least 14 LNs were harvested. All of the patients were pathologically staged according to the 7th AJCC staging system 35   www.nature.com/scientificreports www.nature.com/scientificreports/ acquired 60 minutes after 5.5 MBq/kg (Discovery STe) or 4.0 MBq/kg (Biograph mCT) of FDG was administered intravenously. First, low-dose CT scan (Discovery STe; peak voltage of 120 kVp and slice thickness of 3.75 mm, Biograph mCT; peak voltage of 120 kVp and slice thickness of 3 mm) was acquired for attenuation correction. Immediately following the CT scan, PET scan was obtained with an acquisition time of 3 min per bed position for Discovery STe and 1.5 min per bed position for Biograph mCT in 3D mode. PET images were reconstructed using an ordered-subset expectation maximum iterative reconstruction algorithm. Image Analysis. [ 18 F]FDG PET/CT images were retrospectively interpreted in consensus by two experienced nuclear medicine physicians. First, LN metastases status was visually assessed and categorized into one of two groups. The metastatic LNs were categorized as positive, which showed increased [ 18 F]FDG uptake from the surrounding background activity on PET regardless of size on CT. Subsequently, the volume of interest (VOI) of the primary tumor and LNs were manually drawn, and T_SUVmax and N_SUVmax were measured only in patients with positive [ 18 F]FDG uptake for semiquantitative analyses. We assigned the SUVmax as 0 to patients with negative [ 18 F]FDG uptake of the primary tumor or LNs. The SUVmax was calculated using the following formula: SUVmax = maximum activity in the region of interest (MBq/g)/[injected dose (MBq)/body weight (g)]. T_MTV, obtained with an SUV threshold of 2.5, was used to define the VOI.
Statistical Analysis. Numeric data are expressed as mean ± standard deviation. Univariate and multivariate logistic regression analyses were performed to identify significant variables associated with LN metastasis. The prediction model of LN metastases, with the combination of significant parameters, was developed by using multivariable logistic regression modeling. A nomogram was established to be a graphic representation of the LN metastases prediction model based on the result of multivariate logistic regression analysis. The additional value of risk prediction for LN metastases was evaluated using c-statistics, and DeLong method was used to compare the difference between the AUC 36 . Statistical analyses were performed using MedCalc for Windows, version 18.2.1 (MedCalc Software, Ostend, Belgium) and R version 3.4.3 software (http://www.r-project.org, R Foundation for Statistical Computing, Vienna, Austria). All P values < 0.05 were considered statistically significant.

Data Availability
The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.