Introduction

In the last decade, there has been a sharp increase in colorectal cancer in Korea [1]. This cancer is now the second most common cancer among men and the third most common cancer among women in the country [2]. The surgical resection of the tumor is the most important treatment method in colorectal cancer [3]. In locally advanced rectal cancer, neoadjuvant chemoradiotherapy (NCRT) increases the rate of tumor regression and sphincter preservation [4] and reduces the rate of local recurrence [5]. Therefore, NCRT followed by radical surgery has become the standard treatment method for patients with locally advanced rectal cancer [6].

Patients with the pathologic complete response (pCR) to NCRT generally show better long-term outcomes than those without [7], and only 15–30 % of patients with rectal cancer achieve pCR to NCRT [8, 9]. Although accurately predicting the pCR before radical surgery can facilitate the choice between sphincter-sparing surgery and aggressive surgery [10, 11], the histologic tumor response can be assessed only after surgical resection. Several imaging modalities, such as computed tomography (CT), magnetic resonance imaging (MRI), and endorectal ultrasound (EUS), have been used to predict responses to NCRT in rectal cancer, but these conventional types of morphological imaging modalities show limited accuracy—ranging from 30 % to 60 %—in terms of predicting the tumor response [12]. This limited accuracy is due to inflammation, edema, and fibrosis resulting from NCRT [13, 14].

18F-Fluorodeoxyglucose positron emission tomography (18F-FDG PET) can provide images of viable tumor tissue by reflecting glucose metabolism [15] and produce semiquantitative data [16]. Therefore, 18F-FDG PET has been widely used to diagnose, stage, restage, and monitor treatment responses in many types of malignancies [17, 18]. Many studies have reported the usefulness of 18F-FDG PET for predicting responses to neoadjuvant therapy in rectal cancer [12]. In general, these studies have used various metabolic parameters such as the maximum standardized uptake value (SUVmax), metabolic tumor volume (MTV), total lesion glycolysis (TLG), and changes in PET parameters [19–21], but optimal parameters for predicting the pCR have yet to be defined.

Several studies have reported the usefulness of SUVmax normalized to liver for a semiquantitative assessment of lesions [22–24]. This method can provide reliable data across diverse PET scanners [25] and improve accuracy for the characterization of tumors [26, 27]. Therefore, this study examines the usefulness of PET parameters normalized to liver obtained by 18F-FDG PET/CT before and after NCRT for predicting the pCR in patients with locally advanced rectal cancer.

Materials and Methods

Patient Population

18F-FDG PET/CT images and medical records of 162 consecutive patients with rectal cancer who received NCRT at the authors’ institute from June 2005 to March 2013 were retrospectively reviewed. In this study, patients were enrolled if they met all of the following inclusion criteria: (1) newly diagnosed and histologically proven rectal cancer, (2) the completion of chemotherapy with 5-fluorouracil and leucovorin [11], (3) the surgical resection of the rectal tumor after the completion of NCRT, and (4) 18F-FDG PET/CT before and after NCRT. As a result, those patients who did not undergo surgery (n = 19), those who had no 18F-FDG PET/CT before or after NCRT (n = 54), and a patient who received capecitabine as chemotherapy (n = 1) were excluded, and therefore a total of 88 patients were enrolled in the study.

The Institutional Review Board of the institute approved the study, and no informed consent was required.

18F-FDG PET/CT Acquisition

PET/CT images were acquired on the Biograph6 PET/CT scanner (Siemens Medical Solution, Knoxiville, TN, USA). All patients fasted for at least 6 h before the intravenous injection of 18F-FDG (7.4 MBq/kg of body weight). In patients with a blood glucose level above 7.2 mmol/l, the injection was delayed until the level decreased below 7.2 mmol/l, and the blood glucose level did not exceed 7.2 mmol/l at the time of the 18F-FDG injection in any patient. PET/CT imaging from the skull base to the upper thigh (five to six bed positions) started about 60 min after the injection of 18F-FDG. During the PET/CT scan, CT images with no intravenous iodinated contrast were acquired using a six-slice helical CT scanner (130 kVp, 30 mA, 0.6-s/CT rotation, pitch of 6). Then PET emission images were acquired over the corresponding area with a 16.2 cm axial field of view at 3.5 min per bed position. These CT images were used for attenuation correction, and image reconstruction was performed using a conventional iterative algorithm (ordered-subset expectation maximization, 2 iterations and 8 subsets).

Imaging Analysis and the Determination of Parameters

All PET/CT images were reviewed on e-soft workstations (Siemens Medical Systems, Iselin, NJ). An ellipsoid volume of interest including the whole rectal tumor was drawn by not including the adjacent urinary bladder potentially showing high 18F-FDG uptake. Then the SUVmax and MTV of each dataset were measured. In the case of no discernible focal uptake in the rectum after NCRT, a circular region of interest 2 cm in diameter was drawn at the site corresponding to rectal tumor on PET/CT before NCRT.

MTV was automatically calculated using thresholds of SUV 2.0, SUV 2.5, and SUV 3.0, and outcomes were designated MTV(2.0), MTV(2.5), and MTV(3.0), respectively [21, 28]. To normalize the FDG uptake of the rectal tumor, a circular region of interest (ROI) 3 cm in diameter was drawn in the right lobe of the liver showing homogeneous FDG uptake. In addition, a circular ROI 1 cm in diameter was also drawn in the lumen of descending thoracic aorta to measure FDG uptake of the blood pool [29]. Also, the mean SUV (SUVmean) of the liver and that of the blood pool were measured, respectively. Then the ratio of the SUVmax of the rectal tumor to the SUVmean of the liver (SLR) and the ratio of SUVmax of the rectal tumor to the SUVmean of blood pool (SBR) were calculated, respectively.

The SUVmax, SLR, SBR, and MTV of PET before NCRT were defined as SUV1, SLR1, SBR1, and MTV1, respectively, and the SUVmax, SLR, SBR, and MTV of PET after NCRT, as SUV2, SLR2, SBR2, and MTV2, respectively. Percentage changes in PET parameters were calculated as follows:

$$ \begin{array}{c}\hfill \varDelta \mathrm{SUV}\left(\%\right)=\left(\mathrm{SUV}1-\mathrm{SUV}2\right)\times 100/\mathrm{SUV}1\hfill \\ {}\hfill \varDelta \mathrm{SLR}\left(\%\right)=\left(\mathrm{SLR}1-\mathrm{SLR}2\right)\times 100/\mathrm{SLR}1\hfill \\ {}\hfill \varDelta \mathrm{SBR}\left(\%\right)=\left(\mathrm{SBR}1-\mathrm{SBR}2\right)\times 100/\mathrm{SBR}1\hfill \\ {}\hfill \varDelta \mathrm{MTV}\left(\%\right)=\left(\mathrm{MTV}1-\mathrm{MTV}2\right)\times 100/\mathrm{MTV}1\hfill \end{array} $$

Pathologic Assessment

An experienced pathologist analyzed the surgical specimens containing the primary tumor area and the circumferential resection margin. Pathologic responses to NCRT were classified to two groups. The pCR group was defined as no residual malignant cell other than fibrosis in the surgical specimen (ypT0N0), and the non-pCR group, as any evidence of residual malignant cells in the surgical specimen [30]. According to tumor regression grade (TRG) by Mandard et al. [31], patients with TRG1 were classified as the pCR group, while patients with TRG2-5 were classified as the non-pCR group.

Statistical Analysis

The results are presented as frequency and percentages for categorical variables and the mean ± SD for continuous variables. The Mann–Whitney test was employed to compare PET parameters of the pCR and non-pCR groups. The paired samples t-test was used to compare the SUVmean of liver and blood pool before and after NCRT. The receiver operating characteristic (ROC) curve was analyzed to determine the ability of each parameter to predict the pCR. The pairwise comparisons of ROC curves for PET parameters were performed with the method of Hanley and McNeil [32]. The multi-ROC analysis was performed to determine whether combining PET parameters can improve the diagnostic performance for predicting the pCR [33].

All statistical analyses were conducted using MedCalc 13.0 (MedCalc Software, Belgium). Two-tailed p values of <0.05 were considered statistically significant.

Results

Patient Characteristics

Table 1 shows the characteristics of the 88 patients. The mean patient age was 59.2 ± 11.1, and 73 % were men. The mean distance from the anal verge to the tumor was 4.5 ± 2.7 cm. All patients received a total radiation dose of 5,040 cGy in 28 fractions. The mean interval from PET before NCRT to the initiation of NCRT was 11 ± 6.7 days, and that from the end of NCRT to PET after NCRT was 42.3 ± 7.8 days. The mean interval from the end of NCRT to surgery was 53.2 ± 9.0 days. Based on a histologic examination of surgical specimens, 17 (19 %) patients were classified as the pCR group, and 71 (81 %), as the non-pCR group. The mean values of SUVmean of liver before NCRT and after NCRT were 2.6 ± 0.4 and 2.6 ± 0.5, respectively. The mean values of SUVmean of blood pool before NCRT and after NCRT were 2.0 ± 0.3 and 1.9 ± 0.3, respectively. Also, there was no significant difference in the SUVmean of liver (p = 0.70) and the SUVmean of blood pool (p = 0.65) between before and after NCRT.

Table 1 Patient characteristics (n = 88)
Full size table

A total of 40 (45 %) patients had no discernible focal uptake in the rectum after NCRT, and the mean value of SUV2 measured at the corresponding site of the rectal tumor identified on PET before NCRT was 3.3 ± 1.2 in these patients.

Comparison of PET Parameters Between pCR and Non-pCR Groups

Figure 1 compares the PET parameters between the pCR and non-pCR groups. There were no significant differences in SUV1 (p = 0.89), SLR1 (p = 0.35), SBR1 (p = 0.33), and MTV1 (2.5) (p = 0.14) between the two groups. The pCR group showed significantly lower SUV2 (p = 0.0005), SLR2 (p < 0.0001), SBR2 (p = 0.0001), and MTV2 (2.5) (p = 0.0040) and significantly higher ΔSUV (p = 0.0035), ΔSLR (p = 0.010), ΔSBR (p = 0.044), and ΔMTV (2.5) (p = 0.0015) than the non-pCR group.

Fig. 1
figure 1

Changes of SUVmax (a), SLR (b), SBR (c), and MTV (2.5) (d) after NCRT. Dots indicate median values, and horizontal lines depict interquartile ranges. Red lines indicate the pCR group, and black lines, the non-pCR group. There were significant differences in ΔSUV (p = 0.0035), ΔSLR (p = 0.010), ΔSBR (p = 0.044), and ΔMTV (2.5) (p = 0.0015) between the pCR and non-pCR groups

Full size image

ROC Curve Analysis for Predicting the pCR

The ability of the PET parameters to predict the pCR was calculated using their ROC curves (Fig. 2 and Table 2). Before NCRT, only MTV1 (2.0) was a significant predictor of the pCR (AUC, 0.645; p = 0.029). By contrast, SUV1 (AUC 0.510; p = 0.89), SLR1 (AUC 0.573; p = 0.35), SBR1 (AUC 0.581; p = 0.23), MTV1 (2.5) (AUC 0.617; p = 0.085), and MTV1 (3.0) (AUC 0.602; p = 0.13) were not significant predictors of the pCR. After NCRT, SUV2 (AUC 0.774; p < 0.0001), SLR2 (AUC 0.826; p < 0.0001), SBR2 (AUC 0.815; p < 0.0001), MTV2 (2.0) (AUC 0.724; p < 0.0001), MTV2 (2.5) (AUC 0.724; p < 0.0001), and MTV2 (3.0) (AUC 0.722; p < 0.0001) were significant predictors of the pCR. In terms of percentage changes in the PET parameters, ΔSUV (AUC 0.729; p = 0.0011), ΔSLR (AUC 0.701; p = 0.0060), ΔSBR (AUC 0.664; p = 0.025), ΔMTV (2.0) (AUC 0.687; p = 0.0055), ΔMTV (2.5) (AUC 0.749; p < 0.0001), and ΔMTV (3.0) (AUC 0.713; p = 0.0004) were significant predictors of the pCR. The comparisons of ROC curves for the PET parameters are summarized in Table 3. There were significant differences between SLR2 and SUV2 (p = 0.044), SLR2 and ΔSLR (p = 0.019), SLR2 and ΔSBR (p = 0.0068), SBR2 and ΔSLR (p = 0.0024), SBR2 and ΔSBR (p = 0.0047), ΔSUV and ΔSBR (p = 0.037), respectively. By contrast, there was no significant difference between other PET parameters. Among all PET parameters, SLR2 showed the highest AUC value for predicting the pCR. Based on the Yuden index and the ROC curve, the optimal criterion, sensitivity, specificity, and accuracy of SLR2 for predicting the pCR were ≤1.41, 88.2 %, 64.8 %, and 68.2 %, respectively.

Fig. 2
figure 2

ROC curves of SUV2 (a), SLR2 (b), SBR2 (c), MTV2 (2.5) (d), ΔSUV (e), ΔSLR (f), ΔSBR (g), and ΔMTV (2.5) (h) for the prediction of the pCR. Based on AUC values, all PET parameters predicted the pCR

Full size image
Table 2 PET parameter values for predicting the pCR
Full size table
Table 3 P values for pairwise comparison of ROC curves
Full size table

For the combination of the PET parameters, we selected only SLR2 and ΔMTV (2.5), which showed the highest AUC among the PET parameters after NCRT and percentage changes of the PET parameters, respectively. Among all possible combinations in multi-ROC analysis, combined criterion of SLR2 ≤ 1.01 and ΔMTV (2.5) > 98.1 showed the highest sum of sensitivity and specificity (Fig. 3 in Supplementary Material online). With this criterion, the AUC, sensitivity, specificity, and accuracy for predicting the pCR were 0.759 (p = 0.0007), 58.8 %, 93.0 %, and 86.4 %, respectively. However, it showed no significant difference compared with SLR2 (p = 0.27) or ΔMTV (2.5) (p = 0.90). As a result, the combined PET parameters failed to improve the diagnostic performance for predicting the pCR (Fig. 4 in Supplementary Material online).

Discussion

There are two major findings in the current study. First, PET parameters obtained by 18F-FDG PET/CT before and after NCRT predicted the pCR in patients with locally advanced rectal cancer. Second, among the PET parameters, SLR2 was the most useful parameter for predicting the pCR.

Several imaging modalities have been used for predicting responses to neoadjuvant therapy in patients with rectal cancer, but limited accuracy has prevented the determination of changes in the surgical approach or additional therapy. EUS is an examiner-dependent modality [34], and the reported accuracy of EUS varies from 48 % to 62 % [35]. CT accuracy has been reported to range from 46 % to 65 % [36, 37], and MRI accuracy, from 47 % to 54 % [38–40]. Although the degree of accuracy varies according to the design and setting of the study, these morphological imaging modalities generally overstage the rectal tumor after neoadjuvant therapy [36, 40].

18F-FDG PET has been reported to be more accurate than morphological imaging modalities in predicting responses to neoadjuvant therapy in rectal cancer [37, 41]. Many studies have demonstrated the usefulness of 18F-FDG PET in terms of various metabolic parameters for predicting responses to NCRT in locally advanced rectal cancer [12]. The results of the present study are generally consistent with the findings of previous studies. However, the results are inconsistent in terms of accuracy and optimal cutoff values. These differences may be due the definition of endpoints. In the present study, endpoints were set as the pCR, not as the responder [19, 42, 43], because of the pCR is closely correlated with the local control and better prognosis [11, 44].

SUVmax is a widely used quantitative parameter of 18F-FDG PET, and other volume-based PET parameters such as MTV and TLG are dependent on SUVmax [45]. These PET parameters are limited in terms of a straightforward assessment of metabolic characteristics because various factors such as body weight, obesity, the blood glucose level, and the postinjection uptake time can affect SUVmax [46]. To overcome this problem, normalized SUVmax has been used, and the liver is widely used as a reference region [25]. Several studies have reported the usefulness of 18F-FDG PET for predicting treatment responses by using the tumor-to-liver ratio of FDG uptake in malignant patients [22, 23, 47]. In this study, SLR2 was the best predictor of the pCR. It is conceivable that normalized SUVmax is less affected by above-mentioned factors than other PET parameters. This suggests that SLR2 is not only useful for predicting the pCR but also applicable to various PET scanners.

Among the PET parameters before NCRT, only MTV1 (2.0) predicted the pCR in this study. MTV differs from other PET parameters in reflecting the tumor volume, which is an important independent factor in determining radiotherapy outcome [48]. In addition, MTV represents real tumor burden more accurately than the tumor volume measured by CT [49]. Therefore, the rectal tumor with a larger MTV1 can be more resistant to the standard radiotherapy.

The generally recommended interval between radiotherapy and restaging 18F-FDG PET is at least 3 months to avoid interference by radiation-induced inflammation [50]. However, it is practically inappropriate in a neoadjuvant setting. Many previous studies for rectal cancer performed restaging 18F-FDG PET about 4–6 weeks after neoadjuvant radiotherapy [12]. And there were several reports that inflammation was not a significant problem at this time [41, 51, 52]. In this study, mean interval from NCRT to restaging 18F-FDG PET was 6.1 ± 1.2 weeks, and the shortest interval was 4 weeks. Therefore, we postulated that radiation-induced inflammation had a minimal effect on restaging 18F-FDG PET in the current study.

Although the accuracy of SLR2 for predicting the pCR was superior to other PET parameters, SLR2 plays a limited role in clinical decision making because of its suboptimal accuracy. The gastrointestinal tract can show variable 18F-FDG uptake without pathologic condition [53]. For this reason, physiological 18F-FDG uptake in the rectum may mask residual malignant lesion. In addition, the metabolism of a shrunken rectal tumor after NCRT can be affected by the partial volume effect. These factors can cause false-positive for the pCR, and lead to the relative lower specificities of the PET parameters including SLR2. In this regard, a combination of 18F-FDG PET/CT and other morphological imaging modalities or biochemical markers is expected to improve its accuracy for predicting the pCR in rectal cancer [54, 55].

This study has several limitations. First, the study was designed as a retrospective observational study. Second, survival data on enrolled patients were not considered. Third, it was difficult to promptly apply the suggested optimal cutoff criteria for PET parameters in a clinical setting because they were based on a single institutional data set. Fourth, we did not correct the partial volume effect for the rectal tumor after NCRT.

Conclusions

The results suggest that SLR2 is more useful for predicting the pCR than PET parameters without liver normalization in patients with locally advanced rectal cancer. This new parameter may help clinicians plan surgery and adjuvant therapy.