Patients
Our cohort consisted of 21 chordoma patients from two ongoing National Cancer Institute Institutional Review Board (IRB)–approved phase I clinical trials of therapeutic cancer vaccines. Eleven patients received the yeast-brachyury vaccine GI-6301 (NCT01519817) [11]. Thirteen received MVA-brachyury-TRICOM vaccine (NCT02179515), three of whom had previously received the yeast-brachyury vaccine. CT and MR scans were acquired at baseline and during treatment, and patients who went off trial continued to have follow-up scans. Patients had 2–14 appointments at which imaging was done (median, five). Although all patients had surgery and/or radiation, these treatments were most often given before the baseline scans. Volumetric segmentations were done on subsequent scans up through the most recent scan available for each patient. One patient had to be re-baselined after an ablation for the purpose of our analysis. As a result, that case is used as two separate data points (pre- and post-ablation) for radiologic time to progression (TTP) analysis. Two patients were excluded for not having at least two time points with CT and MR, and another patient was excluded because symptoms recurred after stopping steroids to enroll on a clinical trial. This resulted in three patients who were not included in the Kaplan-Meier analysis and one patient who had two data sets, totaling 19 evaluations. The two patients without follow-up scans were still included in other analyses for the purpose of assessing resources required for volumetric segmentation [12]. This research was conducted on images collected during two clinical trials, which were run in compliance with the Helsinki Declaration and were approved by the Center for Cancer Research, National Cancer Institute Institutional Review Board.
Imaging
CT scans of the chest, abdomen, and pelvis were acquired at baseline (pretreatment) and at 8- to 12-week intervals following treatment initiation using any of the following scanners: Siemens Definition, Biograph, or Flash (Siemens Healthcare USA, Malvern, PA), Toshiba Aquilion ONE™ ViSION CT (Toshiba Medical Systems Corp., Tochigi, Japan), or GE Lightspeed (GE Medical Systems, Waukesha, WI).
Patients received contrast-enhanced CT scans using 0.6- to 2.5-mm collimation, 120 kVp, 150–240 reference mAs (with dose modulation), and 0.25- to 0.75-sec rotation time. Images were pushed to our PACS as contiguous 5 × 5-mm and 2 × 1-mm overlap axial slices for volumetric assessments and reformats (e.g., coronal). Scans were obtained with patients coached to full inspiration, supine from chest to pelvis in one acquisition, and with weight-based (2 mg/kg) i.v. contrast (Isovue 300 at 2 mL/sec) after a 70-sec delay.
One of the following scanners was used to obtain MR scans: 3 Tesla (3 T) Verio (Siemens), 3 T Achieva TX (Philips Healthcare, Andover, MA), 1.5 T Aera (Siemens), 3 T mMR (Siemens), or 1.5 T Achieva (Philips). Patients received TSE T1 axial and coronal imaging, TSE T2 axial and coronal imaging with fat suppression (or STIR), and axial diffusion-weighted imaging with B values of 0, 250, and 800. Apparent diffusion coefficient maps were generated from the 0 and 800 B values. All precontrasted images were acquired at a slice thickness and imaging gap of 6 × 2 mm.
Prior to contrast administration, a precontrast 3D Axial T1-weighted sequence (3-mm overlapping VIBE/DIXON/or E-Thrive) was obtained in a breath-held fashion. Following injection of i.v. gadolinium-based contrast (0.2 mL/kg, injected at 2 mL/s) (Magnevist®, Schering AG, Berlin, Germany and MultiHance®, Bracco, Milan, Italy), postcontrast images were obtained in identical fashion as the precontrast 3D images. Image acquisition time points were 20 sec, 70 sec, and a 3-min delay. All data were automatically subtracted from the precontrast acquisition. A final postcontrast 3D T1-weighted coronal image (3-mm overlapping VIBE/DIXON/or E-Thrive) was obtained at the conclusion of the MR examination.
RECIST measurement
Tumors were evaluated using RECIST 1.1 guidelines [1], which call for one-dimensional, longest-diameter measurements in the axial plane. A maximum of five lesions may be evaluated in each patient, with no more than two per organ system.
Volumetric measurement
A neuroradiologist (NP) reviewed the MR sequences to determine the best ones to use for segmentation. Post-contrast scans were not as useful as expected due to prior radiation and surgical treatments; enhancement was poor and tumors could not be differentiated from adjacent structures. Fat-suppressed T2-weighted and STIR sequences were deemed the most appropriate for segmenting sacral and paraspinous tumors, whereas post-contrast FLAIR sequences were used for clival lesions. Contrast-enhanced CT sequences were used to segment all metastases.
A research assistant (KF) performed the segmentations using the lesion management application within PACS (Vue PACS v 12.0, Carestream Health, Rochester, NY) as previously described [12]. In short, the proprietary software allows the user to identify the edges of the lesion with a digital caliper-like tool and then, based on imaging characteristics, the software generates a proposed border for the lesion across all cuts. To do this, the Vue PACS livewire segmentation tool applies a combination of fast marching [13] and level set [14] algorithms together with shape interpolation for region growing. The cost functions are based on image gradient strengths and image intensity histograms in order to determine the expansion limits. The user (KF) can then correct the border with a correction tool. MR was used for segmenting primary tumors and CT for metastatic disease. Bone metastases were not evaluable by volumetric segmentation. Tumors with long diameters < 0.5 cm were deemed immeasurable due to the inherent variability created by measuring very small lesions, similar to what is outlined in RECIST 1.1. Masses were reviewed and deemed to be measurable tumors based on clinical assessment and imaging characteristics; not all were biopsy-confirmed.
Radiologist review
A neuroradiologist with 30 years of experience (NP) validated volumetric segmentations of primary tumors, and a body radiologist with 20 years of experience (LF) validated segmented metastatic tumors.
Comparison techniques/statistics
Using the following criteria, we divided patients into two groups independent from radiologic analysis for TTP. Patients were placed into either a good or a poor clinical outcome group, based on the presence (poor) of ≥ 1, or the absence (good) of all of the following clinical indicators: (1) increasing tumor-related pain requiring significant change in pain medications, (2) increasing neurologic dysfunction, and/or (3) decreasing ECOG performance status due to tumor-related symptoms [15]. The determination of clinical outcome was made retrospectively at least six months after initial imaging studies.
For patients in each category, Kaplan-Meier curves were used to calculate radiologic TTP by RECIST and by volume (Fig. 1) using the log-rank test for equality of survivor functions. A hazard ratio was also calculated using the Cox proportional hazard regression. TTP by RECIST was assessed using RECIST 1.1, with progressive disease (PD) being an increase of ≥ 20 % in the sum of the longest diameters (SLD). TTP by volume was determined based on previously outlined criteria [16], with PD being an increase of ≥ 40 %. TTP was assessed based on date of enrollment to time of PD by RECIST or volumetric criteria. Patient data were censored if PD criteria were not met on the last imaging studies prior to a local intervention on a target lesion or date of last available imaging.