Background

Breast cancer is the most frequent malignant disease and the fifth leading cause of cancer death in Japanese women. Most of these breast cancers are detected at relatively early stages, and the 5- and 10-year survival rates are reported to be > 90 and 80%, respectively [1]. However, even among stage I or node-negative cases, relapse or distant metastases can occur after initial therapies, and early detection of cases with high recurrence risk would be helpful in improving the overall prognosis of patients with breast cancer.

Conventional modalities for imaging diagnosis comprise mammography, ultrasound, computed tomography (CT), magnetic resonance imaging (MRI), and bone scintigraphy. It was reported that dynamic contrast enhanced MRI and diffusion weighted imaging were correlated with the status of hormone receptors and Ki-67 in primary breast cancer [2, 3]. In recent years, 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT) has come to play an increasing role in the diagnosis of biological properties as well as staging, treatment monitoring of residual disease, and detection of disease recurrence in breast cancer patients [4, 5]. For that purpose, the maximum standardized uptake values (SUVmax) of 18F-FDG has been shown to be correlated with tumor size, nuclear grade (NG), and Ki-67 labeling index (LI) [6, 7]. Furthermore, several studies [8,9,10,11] have shown that the SUVmax of primary tumor, that reflects its metabolic activity, on 18F-FDG PET/CT can predict patients’ poor prognosis.

In malignant tumors, glucose metabolism is usually enhanced, and the uptake of 18F-FDG increases. Therefore, a higher level of 18F-FDG accumulation in PET/CT should reflect higher proliferative activity of the tumor cells. Recently, several studies and meta-analyses have been performed on the relationships between PET/CT and histopathological findings in the field of diagnostic oncology [6, 12,13,14,15,16]. Especially, the uptake of 18F-FDG was shown to be correlated with expressions of histopathological markers, e.g., Ki-67 LI, vascular endothelial growth factor, and hypoxia induced factor 1α, in head and neck cancer, lung cancer, and lymphoma [12,13,14,15,16].

Because the SUVmax is usually measured at a single time point, such as 1 h after 18F-FDG administration, the dynamic index of the tumor is not included in routine examination. Some articles reported the utility of measurement of 18F-FDG uptake levels at dual time points [17, 18]. The 18F-FDG uptake level at a later point has a tendency to increase in malignant lesions but to decrease in benign lesions, such as inflammatory reactions [19]. Therefore, the measurement of 18F-FDG uptake in dual time point 18F-FDG PET/CT may be able to estimate biological properties and predict patient prognosis more accurately.

The aim of this study was to investigate the clinicopathological significance of dual time point 18F-FDG PET/CT in patients with primary breast cancer. In addition, we assessed the prognostic significance of the measurement of dynamic 18F-FDG uptake levels.

Methods

This was a retrospective study in a single institute.

Ethics approval and consent to participate

This study was performed in accordance with the Declaration of Helsinki and was approved by the institutional review board of National Defense Medical College (registration number: 2695). All patients agreed to participate in this study, and written informed consent was obtained from all these patients.

Eligible patients

Between September 2008 and December 2017, 18F-FDG PET/CT was performed for 820 consecutive preoperative patients with primary breast carcinoma. Of these, 356 patients were excluded from the study because of (1) history of malignant diseases other than breast cancer within 5 years, (2) preoperative medication therapy, (3) diabetes mellitus, (4) previous treatment of ipsilateral or contralateral breast cancer, (5) presence of distant metastases, (6) acquisition of only single time point data of 18F-FDG PET/CT, and/or (7) difficulty in measuring SUVmax due to low 18F-FDG accumulation. Finally, 464 female patients were eligible.

In all cases, diagnosis of breast cancer was made based on cytopathological and/or histopathological examination before surgery. 18F-FDG PET/CT was performed before surgery, and the interval between the PET/CT examination and surgery was 42 days on an average, ranging from 7 to 71 days. Postoperative surveillance for 5 years was performed through examination every 3 months and mammography every year. After 5 years, patients underwent mammography every year and were followed up to 10 years after surgery. If relapse was suspected in these tests, it was confirmed using CT or PET/CT.

Quantification of 18F-FDG uptake in primary breast cancer

All 464 patients underwent 18F-FDG PET/CT at the Tokorozawa PET Diagnostic Imaging Clinic (Tokorozawa, Japan). Patients fasted for at least 4 h before the examination. One hour after intravenous administration of 3.7 Mbq/kg 18F-FDG, the first scanning was performed. The first examination involved whole-body imaging from the head to thigh, and the second scanning involved the chest only, within 50–60 min of the first examination.

After image reconstruction, the region of interest (ROI) was placed in primary breast cancer. The SUV is defined as decay-corrected tissue activity divided by the injected dose per patient body and is calculated using the formula,

$$ \mathrm{SUV}=\mathrm{activity}\ \mathrm{in}\ \mathrm{ROI}\ \left(\mathrm{MBq}/\mathrm{ml}\right)/\mathrm{injected}\ \mathrm{dose}\ \left(\mathrm{MBq}/\mathrm{kg}\ \mathrm{body}\ \mathrm{weight}\right). $$

The SUVmax1 and SUVmax2 were obtained at dual time points, and the ΔSUVmax% was calculated using the formula,

$$ {\Delta \mathrm{SUV}}_{\mathrm{max}}\%=\left[\left({\mathrm{SUV}}_{\mathrm{max}}2-{\mathrm{SUV}}_{\mathrm{max}}1\right)/{\mathrm{SUV}}_{\mathrm{max}}1\right]\times 100, $$

where the SUVmax1 and SUVmax2 were the SUVmax at the initial phase (60 min) and SUVmax at delayed phase (120 min), respectively.

Histological study

Two observers (H.T., Y.Y.) performed pathological diagnosis. NG was defined according to the General Rules for Clinical and Pathological Recording of Breast Cancer, 17th edition [20]. NG was determined by the sum of the nuclear atypia score and the mitosis count score. Estrogen receptor (ER) and progesterone receptor (PgR) expression was assessed by immunohistochemistry and defined as positive if ≥1% of carcinoma cells were immunoreactive [21]. Human epidermal growth factor receptor 2 (HER2) positivity was determined according to the American Society of Clinical Oncology/College of American Pathologists guideline 2013 [22]. According to the recommendation of the Breast Cancer Working Group, Ki-67 LI was defined as high if 14% or higher of constituent carcinoma cells were immunoreactive [23, 24]. Pathological stage was determined by the clinical and pathological recording of breast cancer, 8th edition, by Union for International Cancer Control (UICC).

Evaluation of 18F-FDG PET/CT results as prognostic factor

Receiver operating characteristic (ROC) curves were drawn to determine the optimal cutoff values of SUVmax1 and ΔSUVmax%. Furthermore, the Youden index [= sensitivity – (1 – specificity)] of each cutoff value was calculated, and the highest value was taken as the optimal cutoff point.

Statistical analysis according to clinicopathological factors and prognosis

The correlations between SUVmax parameters (SUVmax1, SUVmax2, and ΔSUVmax%) and clinicopathological factors were evaluated using the non-parametric Wilcoxon and the Kruskal–Wallis tests. All statistical analyses were two-sided, with significance defined as P value of < 0.05. The Kaplan-Meier curves for relapse-free survival (RFS) and overall survival (OS) were drawn, and their differences were tested by the log-rank test. A Cox proportional hazards model was used for univariate and multivariate analyses for RFS. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the accuracies of SUVmax1, ΔSUVmax%, and their combination for RFS were calculated. Statistical analyses were performed using JMP® 13 (SAS Institute Inc., Cary, NC, USA).

Results

Patient characteristics

Data obtained from the 464 patients on age, tumor invasion size, histological type, NG, lymphatic invasion, hormonal receptor status, HER2 status, Ki-67 LI, pathological stage, SUVmax1 and SUVmax2, ΔSUVmax%, RFS, and OS are summarized in Table 1. Mean SUVmax1, mean SUVmax2, and mean ΔSUVmax% were 4.6 (± 3.5 standard deviation [SD]), 5.6 (± 4.9 SD), and 15.6% (± 20.2 SD), respectively. SUVmax1 and SUVmax2 did not show normal distribution whereas ΔSUVmax% showed normal distribution (Additional file 1 Figure S1). Five and 10-year RFS rates were 92.0 and 84.9%, respectively. Five and 10-year overall survival rates were 97.3 and 88.5%, respectively (median follow up 4.9 years).

Table 1 Patient characteristics
Full size table

Setting of optimal cutoff values for patient prognostication

According to the Youden index, the optimal cutoff value of SUVmax1 was 3.4, and area under the curve (AUC) was 0.627 (95% confidence interval [CI] 0.536–0.719) (Fig. 1A). The patients were divided into the low SUVmax1 (< 3.4) (n = 223) and high SUVmax1 groups (≥ 3.4) (n = 241). The optimal cutoff value of ΔSUVmax% was 12.5, and AUC was 0.594 (95% CI 0.505–0.683) (Fig. 1B). The patients were divided into the low ΔSUVmax% (< 12.5) (n = 202) and high ΔSUVmax% groups (≥ 12.5) (n = 262).

Fig. 1
figure 1

Determinations of the cutoff point for maximum standardized uptake value at 60 min (SUVmax1) and ΔSUVmax% with reference to relapse events. (a) Receiver operator characteristic (ROC) curves of SUVmax1 for relapse-free survival (n = 464). SUVmax1 at the cutoff value was 3.4, area under the curve (AUC) was 0.627 (95% CI: 0.536–0.719). (b) ROC curves of ΔSUVmax% for relapse-free survival (n = 464). At the ΔSUVmax% cutoff value of 12.5, AUC was 0.594 (95% CI: 0.505–0.683)

Full size image

Patient characteristics between high and low groups divided by SUVmax1 and ΔSUVmax%

The correlations between high and low SUVmax1 groups and clinicopathological parameters are presented in Table 2. Tumor size, pathological T factor, NG, lymphatic invasion, pathological N factor, pathological stage, and SUV parameters (SUVmax1, SUVmax2, ΔSUVmax%) were significantly different between high and low SUVmax1 groups. High Ki-67 LI was more frequent in the high SUVmax1 group than in the low SUVmax1 group (P < 0.0001) whereas ER, PgR, HER2, and subtype were not correlated with SUVmax1. The correlations between the high and low ΔSUVmax% groups and clinicopathological parameters are presented in Table 3. The factors correlated with SUVmax1 were significantly different between the high and low ΔSUVmax% groups. High Ki-67 LI was more frequent in the high ΔSUVmax% group than in the low ΔSUVmax% group (P = 0.0336) whereas ER, PgR and subtype were not correlated with ΔSUVmax%. HER2 status was significantly different between high and low ΔSUVmax% groups (P = 0.0304). Two patients with HER2-positive ductal carcinoma in situ (DCIS) were classified into the low ΔSUVmax% group. Therefore, when these DCIS cases were excluded from the analysis, HER2 status showed no significant difference between these two groups.

Table 2 Patient characteristics between high and low SUVmax1 groups
Full size table
Table 3 Patient characteristics between high and low ΔSUVmax% groups
Full size table

Correlation between SUVmax1 and ΔSUVmax%

There was a weak correlation between SUVmax1 and ΔSUVmax% (P < 0.0001, R2 = 0.166). In the high SUVmax1 group (≥ 3.4) (n = 241), 179 patients (68.3%) with high ΔSUVmax% (≥ 12.5) were included. In contrast, in the low SUVmax1 group (< 3.4) (n = 223), 83 patients (31.7%) with high ΔSUVmax% were included.

Comparison of survival curves

The RFS curves for the high and low SUVmax1 groups were significantly different between these curves (P = 0.0003) (Fig. 2A). Although there was no significant difference in OS curves for the high and low SUVmax1 groups, the high SUVmax1 group tended to show worse prognosis (P = 0.0553) (data not shown). The RFS curves for the high and low ΔSUVmax% groups were significantly different (P = 0.0151) (Fig. 2B). Although, there was no significant difference in OS curves between the high and low ΔSUVmax% groups, the former groups tended to show worse prognosis (P = 0.141) (data not shown). Because the correlation of SUVmax2 with RFS was weaker than that of SUVmax1 (P = 0.0012), we did not use SUVmax2 for prognostic analysis (data not shown).

Fig. 2
figure 2

Relapse-free survival (RFS) curves for (a) patient groups with high and low SUVmax1 values and (b) for patient groups with high and low ΔSUVmax%. (a) RFS curves were significantly different between two patient groups (P = 0.0003). (b) RFS curves were significantly different between two patient groups (P = 0.0151)

Full size image

Prognostication by the combination of SUVmax1 and ΔSUVmax%

The 464 patients were classified into three subgroups (group A, B, and C) by the combination of SUVmax1 and ΔSUVmax%. Group A was SUVmax1 ≥ 3.4 and ΔSUVmax% ≥ 12.5 (n = 179), group B was SUVmax1 ≥ 3.4 and ΔSUVmax% <  12.5 (n = 62), and group C was SUVmax1 <  3.4 (n = 223). Although group C could also be subclassified into the high ΔSUVmax% (n = 83) and low ΔSUVmax% subgroups (n = 140), no significant difference in RFS was observed between these two subgroups (P = 0.625, data not shown).

There were significant differences in RFS curves between these three subgroups (P = 0.0006), and between groups A and C (P = 0.0001). On the other hand, there were no significant differences between groups A and B (P = 0.285), and between groups B and C (P = 0.146) (Fig. 3A). The 10-year RFS rates were 90.6% in group B and 89.0% in group C, whereas the rate was 78.8% in group A. Furthermore, RFS curves were significantly different between group A and group “B + C” (P = 0.0002) (Fig. 3B). By the combination of the ΔSUVmax% and the SUVmax1, it was possible to predict a group with the worse prognosis more sensitively than SUVmax1 or ΔSUVmax% alone.

Fig. 3
figure 3

(a) RFS curves for the patients of subgroups a, b and c classified by the combination of SUVmax1 and ΔSUVmax%. RFS curves were significantly different among these three groups (P = 0.0006). (b) RFS curves for the patients of subgroup A and subgroup “B + C”. RFS curves were significantly different between these two groups (P = 0.0002). Ten-year RFS rates were 78.8% in group A and 89.0% in group “B + C”

Full size image

In the subgroup analyses, there were significant differences in RFS between group A and group B/C in node-negative patients (n = 334) and in node-positive patients (n = 130) (P = 0.0126 and P = 0.0455, respectively). In the pTis/pT1 (n = 297) and pT2/pT3 groups (n = 167), there were no significant differences in RFS between group A and group B/C (P = 0.120 and P = 0.131, respectively). With regard to subtype, group A showed a significantly lower RFS than group B/C in the ER-positive/HER2-negative group (P = 0.0008, n = 345), but such a relationship was not found in the ER-positive/HER2-positive, ER-negative/HER2-positive, and ER-negative/HER2-negative patient groups (P = 0.0614, P = 0.358, P = 0.823, respectively).

Univariate and multivariate analyses

By Cox’s univariate analyses to estimate relapse risk, five clinicopathological parameters, invasive tumor size, lymph node metastasis, NG, lymphatic invasion, and Ki-67 LI, as well as SUVmax1 and ΔSUVmax% were statistically significant factors. The combined SUVmax1 and ΔSUVmax% was also a significant prognostic factor in RFS (P = 0.0007) (Table 4). Because SUVmax1 and ΔSUVmax% were correlated with together, we performed the Cox’s multivariate analyses including these five clinicopathological parameters with either SUVmax1, ΔSUVmax%, or the combination of SUVmax1 and ΔSUVmax%. In the multivariate analyses, SUVmax1 or the combination of SUVmax1 and ΔSUVmax% was an independent prognostic factor (P = 0.0267 and P = 0.0283, respectively, Table 4). As the test to detect relapse, the combined measurement of SUVmax1 and ΔSUVmax% showed higher specificity, PPV, and accuracy than the measurement of SUVmax1 or ΔSUVmax% alone (Table 5).

Table 4 The univariate and multivariate analyses for relapse
Full size table
Table 5 Accuracy of SUVmax1, ΔSUVmax%, and their combination for prediction of relapse
Full size table

Discussion

In malignant tumors, glucose metabolism is usually enhanced, and the extent of increase in glucose consumption was shown to be correlated with higher proliferation rates of cancer cells. Therefore, a higher level of accumulation of 18F-FDG in PET/CT is a sign of the primary breast cancer with high proliferative activities [8,9,10, 25], and 18F-FDG PET/CT has been used not only for cancer diagnosis but also for functional assessments of breast cancer, i.e., clinical aggressiveness and higher sensitivity to neoadjuvant therapies [26, 27]. In fact, Deng et al. and Surov et al. summarized that the uptake of 18F-FDG was associated with Ki-67 LI in their meta-analyses [6, 7]. We were able to confirm their results in this study.

For the evaluation of PET/CT images, the most commonly used parameter is the SUVmax, which is usually measured 60 min after the injection of 18F-FDG. It has also been believed that the addition of information of the later phase can be used to determine the biological properties of the examined cancers in more detail. Some reported that 18F-FDG uptake in malignancy continued to increase until approximately 4–5 h after injection, but the uptake decreased in the benign lesion 30 min after the injection [18, 28]. Furthermore, the ΔSUVmax% was correlated with the grade of malignancy in lung cancer and lymphoma [29, 30]. Although the usefulness of ΔSUVmax% was generally considered acceptable, few reports have been published on its relationship with the prognosis of breast cancer.

In this report, we confirmed that SUVmax1 was an independent prognostic factor for RFS. Furthermore, we showed that ΔSUVmax% was a significant prognostic indicator of RFS and that the combination of SUVmax1 and ΔSUVmax% was possible to predict a group with poorer prognosis more sensitively than SUVmax1 alone. With the optimal cutoff value (12.5 of ΔSUVmax%), the subgroup with better prognosis can be detected among from the high SUVmax1 (≥ 3.4) group. In contrast, the effectiveness of SUVmax1 and ΔSUVmax% for OS could not be demonstrated. In the present patient cohort, follow up period is still short, and the number of events appears too small to analyze the effectiveness of ΔSUVmax% for OS prediction.

The RFS rate of patients with breast cancers of the ER-positive/HER2-negative subtype was significantly lower in the high-SUVmax/high-ΔSUVmax% group than in the other groups (P = 0.0008). SUVmax was shown to be correlated with 21-gene recurrence score in ER-positive/HER2-negative breast cancer [31]. Therefore, SUV-related parameters might be clinically useful, in addition to the 21-gene recurrence score, for the selection of high-risk node-negative luminal breast cancers, although a larger-scale study is necessary. Furthermore, the combination of SUVmax1 and ΔSUVmax% would be able to increase the accuracy of preoperative diagnoses of lymph node metastasis and therapeutic response to neoadjuvant therapies.

In this study, patients with previous treatment were excluded. In these patient groups, 24 ER-negative/HER2-positive patients and 54 ER-negative/HER2-negative patients were included. Therefore, only 10.8% (50/464) were HER2-positive type and 11.6% (56/464) were ER-negative/HER2-negative type. These types of breast cancers were reported to have a higher SUV value than ER-positive types and to have worse prognosis than the ER-positive types [32,33,34]. Furthermore, we excluded the 109 patients whose 18F-FDG accumulation was not visible and SUVmax was not measurable from the study. These cases appear to show very low SUVmax values and accordingly, were also expected to have a good prognosis. For these reasons, it seemed that the true efficacy of ΔSUVmax% and combined measurement of SUVmax1 and ΔSUVmax% as prognostic indicators might be higher than the present results.

The pN factor was a very strong prognostic factor in the univariate analysis but did not have an independent prognostic power in the multivariate analysis. In these analyses, pN was divided into pN0 and pN1–3. Because pN1 was shown to reveal relatively good prognosis and a majority of pN-positive patients showed pN1 in this study, the impact of node-positivity might have been diluted by the good-prognosis effect in pN1 cases. Lymphatic invasion and pT might also have been confounding factors along with pN.

The limitations of this study include its retrospective nature, single center data, and a relatively small number of events. A multicenter, prospective study is needed to highlight the effectiveness of ΔSUVmax% in the prognostication of primary breast cancer.

Nonetheless, the strength of the present study involves the large number of images reviewed, the correlation between relevant clinicopathological and prognostic data, and exclusion of patients with diabetes. Furthermore, SUVmax parameters were easy to compute and reproducible, and dual time point imaging could be performed in a relatively short time with minimal inconvenience to the patient and be readily performed at most centers.

Conclusions

In conclusion, dual time point 18F-FDG PET/CT can be a useful modality for prediction of relapse in patients with breast cancer. The combination of SUVmax1 and ΔSUVmax% was able to identify the patient groups with worse prognosis more accurately than SUVmax1 alone.