Introduction

Osteosarcoma is the most common pediatric malignant bone tumor with an incidence of only a few cases per million person-years on a worldwide basis [1]. The inclusion of aggressive polychemotherapy and the interdisciplinary treatment concept has dramatically improved the long-termsurvival of patients with osteosarcoma. Response to preoperative chemotherapy has been reported to be the most powerful prognostic indicator of survival in osteosarcoma patients [1, 2]. However, histologic response can only be determined in resected specimens, and thus, response monitoring during the course of chemotherapy is not possible [3, 4]. To overcome these limitations, other diverse imaging modalities have been investigated [5].

PET using [18F]FDG can detect changes in tissue metabolism that usually precede structural changes [6], and several studies have shown that FDG PET is useful for noninvasive evaluation of response to neoadjuvant chemotherapy in osteosarcoma [5, 7–11]. Of the various parameters used for measuring tumor glucose metabolic activity, maximum standardized uptake value (SUVmax) provides the most robust measurement in sarcoma [12]. However, the correlation between SUVmax values, which represent only a pixel-sized region of interest (ROI), and histologic response, which averages the percentage tumor necrosis across the entire resected tumor specimen, is not fully understood [8, 13].

In this prospective study, we evaluated the usefulness of SUVmax in predicting and assessing histologic response to neoadjuvant chemotherapy in patients with osteosarcoma. We also analyzed the correlation between FDG PET imaging findings and response to chemotherapy in osteosarcoma using parallel PET and MRI.

Materials and methods

Patient selection

A total of 26 consecutive patients with osteosarcoma treated at our institution were prospectively enrolled between May 2010 and August 2011. Eligibility requirements included biopsy-proven primary high-grade extremity osteosarcoma, the completion of neoadjuvant chemotherapy and surgery, FDG PET/CT and MRI scans obtained before and after neoadjuvant chemotherapy, a time between the first FDG PET scan and the initiation of chemotherapy of no more than 2 weeks, and a time between the second FDG PET scan and surgery of no more than 2 weeks.

All patients underwent a conventional evaluation (plain radiography and MRI of the primary tumor, a 99mTc-methylene diphosphonate bone scan, and a CT scan of the chest) and FDG PET before neoadjuvant chemotherapy. Diagnoses were confirmed based on histologic examinations of tumor tissues obtained by open or needle biopsy, which were performed on average 2 days (range 1–5 days) before the first PET scan.

Our institutional review board approved the study. All patients provided written informed consent, and the study was performed according to the ethical guidelines of our institutional clinical research committee.

Installation and imaging protocol of parallel PET/MRI

A whole-body FDG PET/CT scanner (Biograph6; Siemens Medical Solutions, Malvern, PA) and a 3.0-T whole-body MRI scanner (MAGNETOM Trio A Tim; Siemens Medical Solutions, Erlangen, Germany) were installed in parallel. The two scanners were located 10 m apart separated by an antimagnetic wall. The intravenous administration of FDG was followed by a tracer uptake phase of 60 min. FDG PET/CT scans were obtained 60 min after FDG injection and MRI images were acquired 90 min after FDG injection. Both PET/CT and MRI images were acquired in a supine position, and all patients walked from the PET/CT scanner to the MRI scanner (on independent patient tables). FDG PET and MRI images were coregistered using syngo FusedVision3D software (Siemens Medical Solutions, Erlangen, Germany).

FDG PET/CT

Patients fasted for at least 6 h before intravenous injection of FDG (8.14 MBq/kg). Truncal PET scans were obtained in two-dimensional mode using five to seven bed positions to ensure adequate coverage from the head to the pelvic floor. Additional regional PET scans were also acquired in the same manner as the truncal scans (using three to five bed positions) to cover tumor sites located in the lower extremities. The CT scans were obtained immediately before the PET scans using a six-slice helical CT scanner. The imaging parameters used were as follows: 130 kVp, 30 mA, 0.6 s/CT rotation, and a pitch of 6. No contrast material was administrated. Additional PET/CT scans (six or seven bed positions) were acquired in the same manner as the whole-body scans to cover tumor sites located in the extremities. Emission data were acquired for 3.5 min for each bed position. PET/CT images were reconstructed using the ordered-subsets expectation maximization algorithm (two iterations, eight subsets) with CT-based attenuation correction, after rebinning three-dimensional sinogram using Fourier Rebinning, and smoothed using a 5-mm post-gaussian filter.

FDG PET/CT image interpretation

Abnormal FDG uptake was defined as FDG uptake greater than that in adjacent normal bone. Areas of abnormal FDG uptake were identified, and intensities of FDG uptake were quantified by calculating SUVs from the amounts of FDG injected, total body weight, and regional uptake in attenuation-corrected regional images. Specifically, SUV was defined as the SUVmax of the ROI and calculated from the following equation: (activity/unit volume)/(injected dose/total body weight). PET data were interpreted by an experienced nuclear physician.

MRI

MR images were obtained using a body array surface coil. To ensure the patient’s limbs were in an identical position for PET and MRI, an in-house designed positioning device was used. MRI sequences included a standard T1-weighted sequence (TR/TE 752/10 ms; field of view 450 mm, matrix size 279 × 448; slice thickness 5 mm; 2 average; acquisition time 3 min 53 s), with or without gadolinium enhancement, and a T2-weighted sequence (TR/TE 3,360/97 ms; field of view 450 mm, matrix size 298 × 448; slice thickness 5 mm; 2 average; acquisition time 3 min 7 s), without fat suppression. Gadodiamide (Omniscan; GE Healthcare) was administered intravenously. Intramedullary tumor lengths were measured on coronal sections of unenhanced T1-weighted sequences, and tumor widths and depths were measured on enhanced axial T1-weighted sequences without fat suppression [14]. MR images were independently reviewed by two of the authors. When these two reviewers found a size discrepancy of more than 10 %, images were reviewed simultaneously. Decisions were made in consensus.

Neoadjuvant chemotherapy

All patients underwent two cycles of preoperative chemotherapy using the modified T10 protocol [15]. Briefly, each chemotherapy cycle consisted of high-dose methotrexate, Adriamycin, and cisplatin. Methotrexate at a dose of 8 to 12 g/m2 was administered twice, on days 1 and 7. On day 14, cisplatin was administered at 100 mg/m2 over 2 h. Subsequently, Adriamycin was administered at 60 mg/m2 for 18 h. The intervals between the end of the first cycle of chemotherapy and initiation of the second cycle, and between the end of the second cycle and surgery, were around 3 weeks.

Histologic assessments using PET/MRI images

For detailed specimen mapping, samples were sectioned by an experienced pathologist and an orthopedic surgeon in concert. SUVmax represents the volume of greatest metabolic activity in a tumor, and the physical location of this volume can be ascertained using standard imaging software [13]. In the present study, to locate SUVmax in resected specimens, we applied a technique called mapping. First, the voxel with the SUVmax was selected by adjusting the window of the PET image on the coregistered PET/MRI image. Second, the coronal MRI slice containing the SUVmax was selected and the distance (Z) from the pixel with the SUVmax to the articular surface was measured. Third, the axial MRI slice with the SUVmax was selected and the horizontal and vertical distances (X and Y, respectively) from the pixel with the SUVmax to the axis of long bone were measured. Fourth, one coronal plane Y mm (anterior or posterior) from the axis of the long bone was selected from among initially cut coronal planes with a thickness of 4 mm. Fifth, on this coronal plane, one transaxial plane Z mm from the articular surface was cut at a thickness of 3 mm. Finally, from this stick-like specimen, a 4 × 4 × 3-mm specimen X mm from the axis of the long bone was cut and analyzed (Fig. 1) [13, 16]. Using this method, we were able to determine percentage necrosis by histology at the highest metabolic activity point in a tumor as determined by FDG PET. Because the volume of interest of the SUVmax area was 4 × 4 × 3 mm, we determined a 4 × 4-mm ROI in the corresponding pathologic specimen. This was defined as the minimum histologic response, which can be considered as the percentage necrosis in the region corresponding to SUVmax.

Fig. 1
figure 1

A PET/MRI image was used to locate the point of maximum metabolic activity (SUVmax) in the resected specimen. a The physical location of SUVmax was determined using standard imaging software (arrow). b The resected specimen was initially cut longitudinally in the mid-coronal plane. c Using the premeasured distance from the joint line (determines using a PET/MRI image), the specimen was cut axially. This gross axial section corresponds to a plane containing the SUVmax. d Pathologic mapping was used to locate the point of maximum metabolic activity (SUVmax). The tissue section labeled g contains the highest metabolic activity point in the tumor. e Hematoxylin and eosin staining of the area corresponding to the SUVmax location shows no effect of chemotherapy, which means that the minimum histologic response in this patient was 0 %. The percentage necrosis on histology of the entire resected specimen was 60 % (e ×200)

Full size image

Histologic responses to preoperative chemotherapy were graded using tumor necrosis percentages as grade III and IV (necrosis of 90 % or more), indicating a good response, and grades I and II (less than 90 % necrosis), indicating a poor response [4, 16].

Definition and calculation of parameters

We defined prechemotherapy SUVmax as SUV1 and preoperative SUVmax as SUV2. The SUV change ratio (SCR) was defined as the ratio SUV2 to SUV1. Tumor volumes were determined using MR images as described by [17] using the ellipsoid formula: tumor volume = 0.53 × tumor length × tumor width × tumor depth [14, 18].

Statistics

We analyzed SUV1, SUV2 and SCR in terms of their ability to discriminate responders from nonresponders. For this purpose, we plotted receiver operating characteristic (ROC) curves and calculated areas under the curve (AUCs) for each parameter. We chose parameters that best predicted response and determined cut-off values that showed the highest accuracies. Spearman’s rank correlation test was used to explore correlations between SUV2, minimum histologic response, and histologic response. Analyses were performed using SPSS version 13.0 (SPSS, Chicago, IL). All P values were derived from a two-sided test, and values less than 0.05 were considered significant.

Results

Patient characteristics

Patient characteristics are presented in Table 1. The mean age of the patients was 21 years (range 9–55 years), and 61.5 % were male. The average tumor volume was 128 cm3 (range 20–407 cm3). On the basis of the revised American Joint Committee on Cancer staging system [19], 14 patients (53.8 %) had a stage IIA tumor, 8 (30.8 %) had a stage IIB tumor, 1 (3.8 %) had a stage III tumor, and 3 (11.6 %) had a stage IV tumor. Half of the 26 patients presented with a tumor in the distal femur. The median time between the first PET examination and the initiation of chemotherapy was 3 days and the median time between the second PET examination and surgery was 3 days. All patients underwent limb salvage surgery and in 13 patients (50.0 %) the resected specimens showed a good histologic response to neoadjuvant chemotherapy.

Table 1 Patient characteristics
Full size table

ROC curve analysis of response prediction

ROC curves of SUV1, SUV2, and SCR were plotted to predict histologic response. SUV2 and SCR predicted histologic response and the AUCs were 0.805 and 0.775, respectively (Fig. 2). A SUV2 cut-off of >5 predicted poor responders with a sensitivity of 61.5 %, a specificity of 92.3 %, a positive predictive value of 88.9 %, and a negative predictive value of 70.6 %. A SCR cut-off of >0.5 predicted poor responders with a sensitivity of 84.6 %, a specificity of 69.2 %, a positive predictive value of 73.3 %, and a negative predictive value of 81.8 % (Table 2).

Fig. 2
figure 2

ROC curves of SUV1 (a), SUV2 (b), and SCR (c) were plotted to predict histologic response. SUV2 and SCR were found to predict histologic response

Full size image
Table 2 Predictive values of SUV2 and SCR in 26 patients
Full size table

Correlation among SUV2, minimum histologic response and histologic response

A significant correlation was observed between SUV2 and histologic response (Spearman’s rho correlation coefficient −0.642, P < 0.001; Fig. 3a), and both SUV2 and histologic response were significantly correlated with minimum histologic response, as illustrated in Fig. 3b and c, respectively (Spearman’s rho correlation coefficient −0.515 and 0.911, P = 0.007 and P < 0.001, respectively).

Fig. 3
figure 3

Statistically significant correlations were found between SUV2, minimum histologic response, and histologic response. a SUV2 was correlated with histologic response (Spearman’s rho correlation coefficient −0.642, P < 0.001). b SUV2 was correlated with minimum histologic response (Spearman’s rho correlation coefficient −0.515, P = 0.007). c Histologic response was correlated with minimum histologic response (Spearman’s rho correlation coefficient 0.911, P < 0.001)

Full size image

Discussion

FDG PET can detect changes in tissue metabolism that usually precede structural changes, and thus it should be more accurate for assessing treatment response because it can more correctly identify viable residual tumor [5]. Several studies have shown a strong correlation between reductions in tumor glucose metabolism after neoadjuvant chemotherapy and histologic responses of primary bone tumors [5, 7, 8, 20, 21]. For quantification of tumor glucose metabolism, a variety of methods for SUV measurement have been used: SUVmax in a single pixel, background-corrected values, larger or smaller ROIs, and total lesion glycolysis [22]. Among them, SUVmax in a single pixel which represents the most aggressive portion is most widely used for determining tumor SUV [23]. As a single-pixel value, however, SUVmax is adversely affected by noise. The peak SUV, defined as the average SUV within a small ROI centered on a high-uptake part of the tumor, has been suggested as a more robust alternative [22]. However, quantification of individual tumor response using peak SUV is highly sensitive to the shape, size, and location of the ROI [24]. Hence SUVmax instead of peak SUV was chosen as the metabolic parameter for assessing histologic response in the current study. Moreover, SUVmax measurements in a single pixel are readily reproducible by drawing a volume of interest that includes the entire tumor [25]. Of the various ROIs that are used to measure SUV, SUVmax is being increasingly used in a wide range of human cancers [22, 26–30]. However, for determining the ROI, it is still unclear whether total tumor volume or the maximum metabolically active portion is more important [22].

In the present preliminary study, we evaluated the potential of FDG PET SUVmax as an early predictor of osteosarcoma response to preoperative chemotherapy. In osteosarcoma, previous studies have used two FDG PET scan parameters, namely tumor-to-background ratio (TBR) and SUVmax. Whereas SUVmax represents the point of highest metabolic activity in a tumor, the TBR represents a semiquantitative index of tumor glucose use. In a previous study in sarcoma patients, TBR demonstrated a significantly higher interobserver variability than SUVmax [12], and therefore, we used SUVmax rather than TBR. Furthermore, in four previous studies that used SUVmax for predicting response, it was found that SUV2/SUV1 and SUV2 were correlated with histologic response (Table 3).

Table 3 Previous studies that used SUVmax for predicting response
Full size table

Previous studies have demonstrated a significant association between SUVmax and tumor necrosis [5, 8, 11, 21], which appears a reasonable result because tumor necrosis is the strongest known prognostic factors and high SUVmax values have been shown to correspond to poor survival in osteosarcoma [25, 31]. However, tumor necrosis and FDG PET assessments are quite different. Although these methods evaluate similar biologic features (residual tumor cells after therapy), tumor necrosis is the result of an averaged assessment in a representative plane, whereas FDG PET assessments are based on three-dimensional assessment of maximal remaining tumor activity measured in terms of FDG uptake [31]. Moreover, the location of the SUVmax region cannot always be included during posttherapy histopathologic evaluations of bone specimens [13]. Accordingly, the correlation between SUVmax values and chemotherapy responses in osteosarcoma remains to be validated.

In this preliminary study, we introduced minimum histologic response as an intervening variable to investigate how SUVmax values are correlated with tumor necrosis in osteosarcoma. The locations which represent the point of greatest metabolic activity in tumors can be determined in PET/MRI images using standard imaging software. Since osteosarcoma is a malignancy of bone, its anatomic features can be retained using a technique called mapping [13], which allows the anatomic locations of all submitted tissues to be determined. Thus, we were able to determine the percentage necrosis by histology at the point of greatest metabolic activity in resected specimens (Fig. 1).

One important finding of our study was that more than half of good responders (7/13) showed a good response (≥90 % tumor necrosis) at SUVmax locations (Fig. 3c). In particular, three patients with a totally necrotic tumor in whole specimens showed no tumor cells that appeared viable histologically even at SUVmax locations. This may have been because FDG metabolism also reflects a nonspecific (but bioenergetically intense) inflammatory response and scarring around necrotic tumor [8, 31]. On the other hand, eight poor responders showed no effect of chemotherapy at SUVmax locations. These results imply that response prediction using SUVmax cut-off value provides a useful means of predicting poor responders. In the current study, a SUV2 cut-off of >5 showed relatively poor sensitivity (62 %) in spite of good specificity (92 %) for the prediction of a good histologic response. Therefore, another parameter, such as the combined metabolic/volume index, in combination with SUV2 would be helpful [5, 32].

The present study had several inherent limitations. First, the relationship between SUVmax locations and positions in tissue sections was prone to measurement error. Because this method was a highly educated estimate, we determined the physical location of the greatest metabolic activity in the section with a consensus among nuclear physicians, orthopedic surgeons and pathologists. Second, the degree of inflammation or other active processes which also affect SUVmax was not measured. Third, due to the small number of patients, the potential clinical utility of the SUV2 threshold of 5 for distinguishing treatment responders from nonresponders may be somewhat overstated. Fourth, our suggested SUV cut-off values may differ for different PET scanners.

Conclusion

In summary, a SUVmax of more than 5 after neoadjuvant chemotherapy identified the majority of histologic nonresponders (sensitivity 61.3 %, positive predictive value 88.9 %). In addition, a significant correlation was found between SUVmax and necrosis percentages determined by histology at points of greatest metabolic activity, and tumor necrosis at points of maximum metabolic activity were found to be significantly correlated with histologic response in whole resected specimens.