The objective of this study was to prospectively evaluate dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) as an early imaging indicator of tumor histologic response to preoperative chemotherapy and as a possible prognostic factor for event-free survival (EFS) and overall survival in pediatric patients with newly diagnosed, nonmetastatic osteosarcoma who were treated on a single, multi-institutional phase 2 trial.
Three serial DCE-MRI examinations at week 0 (before treatment), week 9, and week 12 (tumor resection) were performed in 69 patients with nonmetastatic osteosarcoma to monitor the response to preoperative chemotherapy. Four DCE-MRI kinetic parameters (the influx volume transfer constant [Ktrans], the efflux rate constant [kep], the relative extravascular extracellular space [ve], and the relative vascular plasma space [vp]) and the corresponding differences (ΔKtrans, Δkep, Δve, and Δvp) of averaged kinetic parameters between the outer and inner halves of tumors were calculated to assess their associations with tumor histologic response, EFS, and overall survival.
The parameters Ktrans, ve, vp, and kep decreased significantly from week 0 to week 9 and week 12. The parameters Ktrans, vp, and Δkep at week 9 were significantly different between responders and nonresponders (P = .046, P = .021, and P = .008, respectively). These 3 parameters were indicative of histologic response. The parameter Δve at week 0 was a significant prognostic factor for both EFS (P = .02) and overall survival (P = .03).
Osteosarcoma (OS) is the most common malignant bone tumors in children in the United States.1 The current treatment for nonmetastatic OS is based on neoadjuvant chemotherapy to induce tumor necrosis and reduce primary tumor volume and facilitate subsequent tumor resection.2 This strategy of preoperative and postoperative chemotherapy in combination with aggressive surgery has improved long-term survival from 20% to 60% or 70% compared with surgery alone.3-7 However, there is no robust prognostic factor with which to stratify patients who have OS for risk-adapted therapy. Although histologic response (ie, the degree of necrosis induced by chemotherapy before surgery) is the most important prognostic factor for event-free survival (EFS) in patients with OS,8-10 it does not represent a true early prognostic factor, because histologic response cannot be evaluated for patient stratification at presentation before any therapy.
Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is an imaging technique that can be used to measure properties of tissue microvasculature, such as tissue perfusion, capillary permeability, and interstitial volume.11, 12 DCE-MRI images are acquired to monitor the whole process of signal changes before, during, and after intravenous injection of a low-molecular-weight, chelated gadolinium contrast agent. Regions of necrosis, muscle, vessel, and viable tumor display distinct signal enhancement in dynamic images. DCE-MRI has been used for a range of clinical applications, including cancer detection,13, 14 diagnosis,15, 16 staging,13, 17 and assessment of treatment response.18, 19
In a small group of patients with OS, DCE-MRI demonstrated potential as a biomarker for histologic response to preoperative chemotherapy.18 DCE-MRI in combination with tumor size also was identified as a possible prognostic factor for pediatric patients with OS.20 However, these parameters were measured after preoperative chemotherapy instead of at presentation, and DCE-MRI parameters alone were not prognostic of clinical outcome.20 Several other imaging modalities, such as diffusion-weighted MRI,21-2518F-fluorodeoxy-D-glucose–positron emission tomography,26-28 and thallium 201 (TI-201) scintigraphy,29 also play important roles in the assessment of treatment response in solid tumors, including OS. However, to our knowledge, no parameter has been reported as a prognostic factor for the outcome of patients with OS. Recently, central tumor photopenia on TI-201 scintigraphy studies of primary OS was associated negatively with survival in older pediatric patients with OS.30 Because central tumor photopenia may be caused by central necrosis,30 the authors of that report hypothesized that the differences in averaged DCE-MRI parameters between outer and inner tumor may be possible indicators of response or prognostic factors for patient outcome.
In the current study, DCE-MRI data from pediatric patients with OS who were treated on a multi-institutional trial were analyzed to generate quantitative measures: the influx volume transfer constant (Ktrans), the efflux rate constant (kep), the relative extravascular extracellular space (ve), and the relative vascular plasma space (vp) from a 2-compartment pharmacokinetic model11 and the corresponding differences (ΔKtrans, Δkep, Δve, and Δvp) between outer and inner tumor. We investigated the hypotheses that 1) quantitative DCE-MRI measures will be indicative of preoperative treatment response to neoadjuvant therapy, and 2) early measures before any therapy will be prognostic of EFS and overall survival.
MATERIALS AND METHODS
Patients and Treatment
In total, 77 patients with high-grade, nonmetastatic, and potentially resectable OS were enrolled on a phase 2 therapeutic trial at 3 centers in United States and Chile between May 1999 and May 2006 (National Clinical Trial NCT00145639; available at: http://www.clinicaltrials.gov [access date October 11, 2011]). All previously untreated patients aged <25 years were enrolled. Five patients were deemed ineligible after enrollment because of the presence of metastatic disease at diagnosis and were excluded. Of the remaining 72 eligible patients, 1 was diagnosed with malignant fibrous histiocytoma at resection and was excluded, resulting in a total of 71 patients (median age, 13.5 years at diagnosis). Protocol treatment was comprised of 12 cycles of chemotherapy administered every 3 weeks over 35 weeks: 3 cycles of carboplatin and ifosfamide and 1 cycle of doxorubicin before surgical resection at week 12 followed by 2 additional cycles of carboplatin and ifosfamide, 3 cycles of ifosfamide and doxorubicin, and 3 cycles of carboplatin and doxorubicin.31
Patients were eligible for the DCE-MRI imaging studies if they completed at least 1 of 3 serial DCE-MRI examinations before surgical resection. Two patients did not meet this criterion, resulting in a total of 69 patients in the study (according to tumor location: femur, 45 patients; tibia, 17 patients; humerus, 3 patients; fibula, 2 patients; ulna, 1 patient; and maxilla, 1 patient). Adequate renal function, which was defined as a serum creatinine level <2 times the normal value, was an eligibility requirement for enrollment on the trial. No renal function requirements for DCE-MRI were specified in the protocol; contrast was administered according to the policies and procedures of the individual participating institutions. The schedule diagram of treatment and imaging is provided in Figure 1. Treatment and imaging protocols were approved by the institutional review boards of the participating institutions, and written informed consent was obtained from the patient, parent, or guardian, as appropriate.
Evaluation of Response
Histologic response was assessed 12 weeks after definitive surgery using the 4-grade system described by Huvos and colleagues.3, 5 Responders were defined by a percentage of chemotherapy-induced necrosis ≥90% (grade 3, 90%-99%; grade 4, 100%), and nonresponders were defined by a percentage of chemotherapy-induced necrosis <90% (grade 1, 0%-49%; grade 2, 50%-89%).32 In addition, patients who were diagnosed early progressive disease before surgery were considered nonresponders for the purposes of statistical analysis.
Three serial DCE-MRI studies were obtained at week 0 before any treatment (N = 62), at week 9 (N = 60), and at week 12 (N = 51) to measure properties of the tumor microvasculature before definitive surgery, as indicated in Figure 1. DCE-MRI images were acquired on a 1.5-Telsa Siemens Symphony scanner (Siemens Medical Solutions, Erlangen, Germany) with the standard quadrature body coil as transmitter and receiver. After selection of the single slice that best revealed the tumor, images were acquired before, during, and after bolus injection into a central venous access of a 0.1 mmol/kg dose of gadolinium-diethylene triaminepenta acetic acid )Gd-DTPA), followed by a saline flush. Thirty sequential fast low-angle shot (FLASH) images (TR/TE=23/10 msec, 40° flip angle; Nx/Ny=256/256; 10 mm thickness; 40-50 cm field of view, 2 acquisitions) were collected over a 6-minute period, providing a temporal resolution of approximately 12 seconds per image.
After DCE-MRI images were transferred to an offline workstation, a pediatric radiologist (F.A.H.) used an interactive display to select the region of interest (ROI) that encompassed the tumor area identified on routine clinical imaging studies and ensured that tumor boundary selection was consistent across all time points. DCE-MRI data were analyzed using a 2-compartment pharmacokinetic model,11 which required an arterial input function (AIF) and mapping of the baseline spin-lattice relaxation time (T10). A measured AIF was not available for these patients; therefore, an assumed AIF biexponential decay curve33 was used. Because T10 mapping was not acquired for all patients with OS, DCE-MRI kinetic parameters were calculated for all patients using an average T10 of 1100 msec. We computed this average from tumor regions in measured T10 maps from 20 patients with OS that were obtained using an inversion-recovery method with 6 different inversion times (100 msec, 300 msec, 900 msec, 1500 msec, 2200 msec, and 3300 msec) and a 3-parameter fitting algorithm.34 It has been demonstrated previously that pharmacokinetic modeling using a population-based, averaged, constant T10 may generate results comparable to those generated using a measured T10 map when the DCE-MRI parameters are averaged for the tumor.35-37
For each pixel inside the tumor ROI, the 4 quantitative measures (Ktrans, kep, ve, and vp) were computed using the 2-compartment pharmacokinetic model, and the average values for the whole ROI were calculated from parametric maps. The reproducibility of DCE-MRI measures has been demonstrated previously in adults who underwent MRI scans daily for 3 consecutive days.38 The 95% confidence interval for change as a percentage of the group mean pretreatment value was −10.8% to 12.1% for Ktrans, −9.5% to 10.5% for kep, and ±5.1% for ve, respectively. The corresponding differences (ΔKtrans = Ktrans[outer] − Ktrans[inner], Δkep, Δve, or Δvp) of each averaged kinetic parameter between the outer half and the inner half of the tumor ROI also were computed for further statistical analysis. All 8 DCE-MRI parameters were used to assess treatment response, EFS, and overall survival.
The average values of each of the 8 DCE-MRI parameters (Ktrans, kep, ve, vp, ΔKtrans, Δkep, Δve, and Δvp) in the ROI were determined for each patient at each time point of examination (week 0, week 9, and week 12). EFS was defined as the time from the date of study enrollment to the date of the first event (relapsed or progressive disease, second malignancy, or death from any cause) or to the date of last follow-up for patients without events. Overall survival was calculated from the date of study enrollment to the date of either death from any cause or the last follow-up.
Exact Wilcoxon signed-rank tests39 were used to examine the association of each of the DCE-MRI parameters between 2 time points. Logistic regression40 was used to examine the association of each of the 8 DCE-MRI parameters at each time point between responders and nonresponders. Cox proportional-hazards models41, 42 were used to explore associations between outcome (EFS and overall survival) and each of the 8 DCE-MRI parameters. All statistical analyses were performed using SAS software (version 9.1; SAS Institute, Inc., Cary, NC).
For survival analyses, patients were categorized into 2 groups using the median DCE-MRI parameter value as a cutoff point. EFS distributions were estimated using the method of Kaplan and Meier,43 and differences in EFS distributions were examined using the exact log-rank test.44 Reported P values were considered statistically significant when P ≤ .05 and were considered marginally significant or trending toward significance when .05 < P ≤ .10. No adjustments were made for multiple comparisons.
Tumor ROIs drawn by the radiologist were divided by computer software into inner and outer halves, as indicated by the black line in Figure 2. Parametric maps (Ktrans and ve) of 2 pediatric patients with OS of the distal femur at the baseline examination (week 0) are displayed as examples. The first patient in the top row was a responder who was event-free, and the second patient in the bottom row was a nonresponder who died after disease relapse. For the first and second patients, the average Ktrans values for the whole ROI were 0.251 and 0.215, respectively; and the average ve values were 0.178 and 0.231, respectively. The Ktrans differences (ΔKtrans) between the outer and inner halves were −0.005 and 0.124, respectively; and the ve differences (Δve) were 0.018 and 0.138, respectively, for the 2 patients. According to the Ktrans and ve maps provided in Figure 2, the first patient had a more highly perfused central tumor region than the second patient, which means presumably enhanced drug delivery for the central tumor for the first patient. Conversely, the second patient had a large, seminecrotic and necrotic region in the central tumor, and drug delivery to the central tumor would be more challenging.
All 8 DCE-MRI parameters were evaluated for all patients at each time point, and average values within the ROI were assessed and are provided in Figure 3. Bar plots of average values for Ktrans, ve, vp, and kep are provided in Figure 3a, and bar plots of average values for ΔKtrans, Δkep, Δve, and Δvp are provided in Figure 3b. In Figure 3, Ktrans, kep, ve, vp, and Δkep values between weeks 0 and 9 and between weeks 0 and 12 differed significantly (all P < .0001, except for P = .015 [ve] and P = .0004 [vp] for week 0 vs week 12). In Figure 3a, the significant decreases in Ktrans, ve, vp, and kep from week 0 to week 9 and from week 12 can be observed. Changes in ΔKtrans, Δve, and Δvp were not significantly different for week 0 versus week 9 or for week 0 versus week 12. In addition, no significant differences in any parameters were observed between weeks 9 and 12 (P ≥ .19).
Tumor Histologic Response
Patients were categorized into 2 groups, responders and nonresponders, according to their histologic tumor response. The association of each of the 8 parameters between the 2 groups was examined using the logistic regression method.40 Figure 4 provides bar plots of Ktrans and vp for responders and nonresponders at each time point. Ktrans and vp at week 9 differed significantly between the 2 groups (P = .046 and P = .021, respectively). Ktrans and vp at week 12 (P = .08 and P = .07, respectively) were marginally significant. No differences in Ktrans and vp at week 0 were observed between the 2 groups (P > .89). No statistically significant differences in kep and ve between responders and nonresponders were observed at any time point. Figure 5 provides bar plots of ΔKtrans and Δkep for responders and nonresponders at 3 time points. Δkep at week 9 was significantly different between the 2 groups (P = .008). ΔKtrans at week 9 was marginally significant (P = .061). No other significant differences were observed for ΔKtrans or Δkep. In addition, no significant differences in Δve and Δvp between responders and nonresponders were observed at any time point. Ktrans, vp, and Δkep at week 9 provided early indicators of histologic response.
Event-Free and Overall Survival
Associations between EFS and each of the 8 parameters were examined using Cox proportional-hazards models.41, 42 ΔKtrans and Δve were 2 parameters that had possible prognostic significance in univariate analyses (Fig. 6a,b). ΔKtrans at week 12 was a significant predictor of EFS (P = .030), and ΔKtrans at week 0 trended toward significance (P = .064). Δve at weeks 0 and 9 was a significant predictor of EFS (P = .002 and P = .040, respectively), whereas Δve at week 12 was marginally significant (P = .070). None of the other 6 parameters were significant predictors of EFS.
To explore the prognostic effect of ΔKtrans and Δve, we plotted EFS curves in Figure 7 for patients who had parameter values above and below the median. Seven patients who did not have DCE-MRI parameter values available at week 0 were excluded from this analysis. Figure 7a provides EFS curves for the 2 groups stratified by the median value of ΔKtrans. EFS was better for patients who had lower values of ΔKtrans at week 0, and this difference was marginally significant (P = .059). Figure 7b provides EFS curves for the 2 groups stratified by the median value of Δve; patients who had lower Δve values at week 0 had significantly better EFS (P = .039). ΔKtrans and Δve at week 0 were potential prognostic factors for EFS before any treatment. To further test the performance of Δve at baseline as a predictor of EFS, a receiver operating characteristics (ROC) curve was evaluated and is illustrated in Figure 8. The area under the curve (AUC) was 0.701, and the optimal cutoff point for best sensitivity and specificity was a Δve value of 0.032, which corresponded to 0.68 sensitivity and 0.70 specificity.
Associations between overall survival and each of the DCE-MRI parameters also were explored. We observed that ΔKtrans and Δve were 2 parameters with possible prognostic significance (Fig. 9a,b). ΔKtrans at week 12 was marginally prognostic of survival (P = .052), although no differences at weeks 0 and 9 were observed (P > .37). Δve values at weeks 0, 9, and 12 were significant predictors of survival (P = .003, P = .048, and P = .036, respectively). Patients who were alive at the time of analysis had smaller average Δve values (Fig. 9b). None of the other 6 parameters were significant predictors of overall survival.
In this study, we examined the relation between DCE-MRI parameters and treatment outcomes (histologic response, EFS, and overall survival), and we demonstrated that Ktrans, vp, and Δkep at week 9 were correlated significantly with histologic response. However, the vp values in this study were very small using the 2-compartment model and may have been affected by the larger noise. No other parameters had statistically significant associations with histologic response. Δve at week 0 was associated significantly with both EFS and overall survival and was the only statistically significant prognostic factor for these clinical outcomes before any treatment. ΔKtrans at week 0 trended toward significance for an association with EFS. ΔKtrans and Δve at later time points also were prognostic factors for EFS. All DCE-MRI parameters that had significance in the univariate test are summarized in Table 1.
Table 1. Summary of Statistically Significant Associations Between Dynamic Contrast-Enhanced Magnetic Resonance Imaging Parameters and Response or Outcomes
Abbreviations: Δkep, corresponding difference in the efflux rate constant; ΔKtrans, corresponding difference in the influx volume transfer constant; Δve, corresponding difference in the relative extravascular extracellular space; EFS, event-free survival; Ktrans, influx volume transfer constant; vp, relative vascular plasma space; wk, week.
.002 (wk 0), .040 (wk 9)
.003 (wk 0), .048 (wk 9), .036 (wk 12)
An observation from Table 1 is that the DCE-MRI parameters that were correlated significantly with histologic response and EFS (or overall survival) were different, and there was no overlap between the 2 groups. For example, Ktrans at week 9 was correlated significantly with histologic response, but it was not a significant predictor of EFS or overall survival. The group of patients with events in the analysis of EFS did not necessarily equate to the group of patients who were nonresponders. In analyses investigating whether tumor response was prognostic of EFS and survival, we observed that histologic response was not a statistically significant predictor of survival (P = .19) or EFS (P = .09) at the traditional P ≤ .05 level. In this analysis, response was treated as a time-dependent covariate; all patients began in the nonresponse state, and patients moved to the response state at the time of their response. Associations between DCE-MRI parameters and histologic response, therefore, may not be equated with associations between the same DCE-MRI parameters and EFS (or overall survival). Each indicator of response or prognostic factor of survival must be validated independently.45
A true prognostic factor for EFS at the time of presentation is most desirable to stratify patients with OS for designing individual treatment strategies. ΔKtrans and Δve values (differences in parameters between outer and inner tumor) at presentation had the potential to be useful prognostic factors for EFS, as indicated in Figure 7. Both ΔKtrans and Δve in patients who had events were greater than those in patients without events, as illustrated in Figure 6, which indicates that patients who had events had a larger drop of perfusion from the outer half of the tumor to the inner half of the tumor than patients without events. Possible reasons for lower perfusion in the tumor central region are central tumor necrosis or high perfusion pressure in the central tumor. Either reason would decrease the drug delivery to the central tumor region and would diminish the effects of chemotherapy, which could lead to less effective treatment and subsequent tumor relapse. Patients who had events usually had much lower perfusion in the central tumor region than at the tumor edge compared with patients who did not have events, as indicated in Figures 2 and 6, which may be a major reason for their different clinical outcomes.
DCE-MRI parameters potentially may be used to incorporate early changes in therapy and in some patients who cannot undergo surgery to assess histologic response. However, the list of active agents in OS is short, and the question whether altering therapy for poor responders improves patient outcome is not yet resolved. Currently, the presence or absence of metastasis at diagnosis is the prognostic factor most widely used to design treatment protocols for OS, and there is a pressing need to identify prognostic factors, especially for the majority of patients (those with localized disease). If these results are confirmed in a larger trial, then the prognostic significance of the DCE-MRI parameters will be useful for stratifying patients in future clinical trials, identifying those patients who warrant the testing of novel therapies, and avoiding the exposure of patients who have a favorable prognosis to potentially toxic or ineffective therapies.
An advantage of ΔKtrans and Δve as prognostic factors for EFS is that these 2 parameters are stable, because they are calculated from a single measurement, which enables the avoidance of effects from slice position changes and motion between measurements. Another advantage is that these parameters can be acquired before any treatment and serve as true early prognostic factors for EFS. However, there are some limitations to this study. Although the acquisition of accurate T10 is preferable for kinetic modeling, these T10 maps were not available for all patients with OS in this study; and a measured, average T10 from a limited sample of patients was used in data processing. Others have demonstrated that this approach yields average kinetic parameters that are not significantly different from those obtained by using a measured T10 map in whole tumor analyses36; however, to our knowledge, the impact of using an average T10 on differences in kinetic parameters between outer and inner tumor halves has not been assessed. Although the results using a measured T10 map may differ from those reported in this study, DCE-MRI examinations processed under the same assumptions should yield comparable results. Another limitation of the current study was that all DCE-MRI parameters were acquired from a single, 2-dimensionanl slice through the tumor for each patient. Although slice positioning was selected carefully based on previous examinations, differences in slice orientation and position may cause increased variation in the results between examinations, but these differences should not have an impact on the observed relations between single examinations and response or survival. A new, 3-dimensional DCE-MRI protocol with 3-dimensional T10 mapping been designed and implemented into an ongoing new clinical trial (National Clinical Trial NCT00667342; available at: http://www.clinicaltrials.gov; [access date October 11, 2011]) to prospectively validate these prognostic factors.
In conclusion, we observed that the DCE-MRI parameters Ktrans, vp, and Δkep at week 9 could serve as indicators of histologic response. The DCE-MRI parameter Δve at week 0 and possibly the parameter ΔKtrans at week 0 may be true early prognostic factors for EFS and overall survival and eventually may contribute to the development of risk-adapted therapy. Further studies with larger numbers of patients are needed to verify our findings and to establish the role of DCE-MRI in stratifying patients for individualized treatment and monitoring their response to chemotherapy.
We acknowledge the valuable contributions of Rhonda Simmons; Gaston K. Rivera, MD; and Dana Hawkins, RN, BSN, CCRC. We thank David Galloway for editorial assistance.
This work was supported in part by Cancer Center support grant P30 CA21765 and Solid Tumor Program Project grant P01 CA23099 from the National Cancer Institute, Bethesda, MD and by the American Lebanese Syrian Associated Charities (ALSAC), Memphis, TN.