Monitoring Response to Antiangiogenic Therapy in Non–Small Cell Lung Cancer Using Imaging Markers Derived from PET and Dynamic Contrast-Enhanced MRI

Study Design and Eligibility Criteria

All subjects included in a prospective multicenter phase II trial evaluating the efficacy of BE treatment in chemonaïve patients with advanced NSCLC were studied using an extensive imaging protocol including CT, DCE MRI, H₂¹⁵O PET, and ¹⁸F-FDG PET. Patients were included when at least 2 imaging studies (including CT) were available at baseline. The protocol was approved by the institutional medical ethics review board of each participating center. Written informed consent was obtained from all patients. Inclusion criteria were histo- or cytologically documented stage IIIB (malignant pleural effusion) or IV nonsquamous NSCLC; no prior systemic therapy; measurable disease, as defined by the response evaluation criteria in solid tumors (RECIST); and an Eastern Cooperative Oncology Group performance status of 0–2. Major exclusion criteria were evidence of the tumor invading major blood vessels; the presence of a cavitating lesion; radiotherapy within 28 d before registration; evidence of bleeding diathesis, coagulopathy, or history of grade 2 or greater hemoptysis; and brain metastasis or spinal cord compression, unless previously treated, with evidence of stable disease for at least 2 mo.

Study Treatment

Patients were treated with bevacizumab (15 mg/kg) as an intravenous infusion every 3 wk and erlotinib (150 mg orally) daily. No dose reductions were made for bevacizumab. In the case of severe side effects, the dose of erlotinib was reduced according to the label. Patients remained on treatment until disease progression, unacceptable toxicity, or patient refusal. In the case of documented tumor progression (as assessed by RECIST), patients received further treatment as per investigator decision. Study medication and imaging were discontinued in these cases.

Imaging Schedule and Parameters

Imaging with CT, DCE MRI, H₂¹⁵O PET, and ¹⁸F-FDG PET was performed at baseline and after 3 wk of treatment (just before bevacizumab infusion) to derive measures on tumor size, vascular permeability, perfusion, and metabolism. Additional CT scans were made every 6 wk from baseline until disease progression. Size was measured with CT and response defined by RECIST. Perfusion (F) was determined with H₂¹⁵O PET; glucose metabolism (metabolic rate of glucose [MR_glu] and the standardized uptake value [SUV]) with ¹⁸F-FDG PET; and a combined measure of microvascular flow, permeability, and surface area (the endothelial transfer constant [K^trans]) with DCE MRI.

Dynamic ¹⁸F-FDG and H₂¹⁵O PET studies were performed selectively at 1 center, whereas the other centers applied a static whole-body ¹⁸F-FDG PET protocol without perfusion measurements. CT and DCE MRI were available for all centers.

Imaging Acquisition

PET was performed with 4 scanners: 2 Siemens ECAT EXACT HR+ scanners, 1 Philips Gemini TF-64 PET/CT scanner, and 1 Siemens Biograph bismuth germanate PET/CT scanner. Dynamic scans were obtained on a single Siemens ECAT EXACT HR+ scanner and static ones on the other systems. All patients were asked to fast for 6 h before scanning. Patients received a venous injection of ¹⁸F-FDG (179–458 MBq), depending on individual patient and scanner characteristics.

Dynamic acquisition started with a 10- to 15-min transmission scan to correct for photon attenuation (13). After a bolus injection of H₂¹⁵O (1,100 MBq), a dynamic emission scan (in 2-dimensional acquisition mode; total duration, 10 min) was started. After an interval of 10 min to allow for decay of H₂¹⁵O, a bolus injection of ¹⁸F-FDG (370 MBq) was administered, and a dynamic emission scan (2-dimensional; total duration, 60 min) was started. Data were reconstructed as 128 × 128 matrices using filtered backprojection with a Hanning filter (cutoff, 0.5 cycle/pixel). For volume-of-interest (VOI) definition purposes, the last 3 frames of the ¹⁸F-FDG sinograms (45–60 min after injection) were summed. This summed frame was reconstructed using ordered-subset expectation maximization (OSEM) with 2 iterations and 16 subsets, followed by postsmoothing of the reconstructed image using a gaussian filter (5 mm in full width at half maximum) to obtain the same resolution as for the filtered backprojection images.

Static acquisition was performed with the following settings: Siemens HR+: 2-dimensional mode, 4-min emission and 3-min transmission per bed position, and OSEM reconstruction (2 iterations, 16 subsets); Philips Gemini: 3-dimensional mode, 135-s emission per bed position, and time-of-flight OSEM reconstruction using default settings; and Siemens Biograph: 3-dimensional mode, 5-min emission per bed position, and Fourier rebinning plus OSEM reconstruction (4 iterations, 8 subsets). All appropriate corrections for normalization, dead time, random coincidences, scatter, and attenuation were applied. Attenuation correction for the PET/CT systems was based on a low-dose CT scan acquired during tidal breathing and on transmission scans with rotating ⁶⁸Ge rod sources for the Siemens HR+ system. All reconstructions resulted in an image resolution of 7 mm at the center of the field of view (FOV).

DCE MRI was performed on three 1.5-T clinical MRI systems: 2 Siemens Sonata systems and an Intera Philips system. All DCE MR images were acquired in a transverse plane (5 slices; slice thickness, 10 mm), with the patient using a breath-hold technique. The acquisition protocol included 5 precontrast T1-weighted (3-dimensional spoiled gradient-echo sequences) measurements with different flip angles (35°, 25°, 15°, 10°, 8°, 4°, and 2°) to determine the T1 relaxation time in the blood and tissue before contrast arrival. Next, gadolinium–diethylenetriaminepentaacetic acid (0.1 mmol/kg of body weight) (Magnevist, 0.5 mol/L; Bayer Schering Pharma) was intravenously injected (3.0 mL/s) and flushed with 15 mL of saline. This injection was followed by the DCE series, using the same sequence as the 5 precontrast T1-weighted measurements but with a flip angle of 35°, containing 30–35 scans of 2 s each. Images were acquired with the following parameters: first Siemens scanner: repetition time/echo time, 2.84/1 ms; FOV, 350 mm; and matrix, 263 × 350, and second Siemens scanner: repetition time/echo time, 2.05/0.75 ms; FOV, 350 mm; and matrix, 160 × 256. For the Philips scanner, parameters were repetition time/echo time, 4.5/2 ms; FOV, 350 mm; and matrix, 144 × 256. A fast data acquisition period (interscan interval, 2 s) was started before the bolus arrival, with the patient under breath-hold instructions to minimize motion artifacts during the first passage of the contrast agent.

Data Processing

During the study, clinicians were unaware of the imaging results, and those who analyzed the images were unaware of clinical outcome and the alternative modalities (except for appropriate FOV positioning of initial PET using the baseline CT).

PET

Patients were evaluable only when all scans were obtained on the same scanner, with a consistent protocol with respect to injected ¹⁸F-FDG dose, interval from injection to scanning, and acquisition settings (i.e., exclusion of within-patient variability) (14). Because individual medical centers used different PET scanners, only the relative change of imaging parameters was used. In this way, heterogeneity due to the use of multiple scanners was nearly eliminated (15). All images were converted to ECAT 7 format and analyzed using software developed in-house. Threshold-defined VOIs of the primary tumor (41% of the maximum pixel value, with correction for physiologic uptake in the local background) were defined semiautomatically (15,16). For dynamic scans, the ¹⁸F-FDG tumor VOIs were applied to both ¹⁸F-FDG and H₂¹⁵O dynamic data to generate time–activity curves. An image-derived input function was created by multiple 2-dimensional regions of interest (ROIs) manually drawn over the aortic arch and ascending aorta. Three venous blood samples were taken as a quality control for the image-derived input function (17). ROIs were then applied to all frames to generate an input time–activity curve. Full kinetic analysis to derive values of the MR_glu was performed using a 3-compartment model and Patlak graphical analysis (18). For all scans, simplified semiquantitative measures were derived by calculating the mean SUV for primary tumor VOIs (50–60 min after injection for dynamic scans and 80 min after injection for static ones) and corrected for lean body mass. The correlation between simplified (SUV) and full kinetic (MR_glu) analysis was explored to validate the use of SUV during this intervention (19). Mean SUV was chosen above maximum because of superior reproducibility (20) and because it allowed the validation of SUV against MR_glu, which cannot be done for a single voxel.

Tumor perfusion was estimated using dynamic H₂¹⁵O data and a 1-tissue-compartment model, considering an arterial blood volume component, and an image-derived input function (21,22).

DCE MRI

A single central slice containing the primary tumor was selected, and an ROI was manually drawn around the whole tumor. Before an ROI was drawn, the unenhanced and enhanced images were both viewed so that nontumor tissue and large vessels were avoided. Care was taken to ensure that tumor ROIs of subsequent scans were drawn at the same anatomic position. To obtain an arterial input function in every patient, an ROI was manually drawn in a major blood vessel using 1 transverse plane in the peak arterial enhancement phase of imaging. Unfortunately, it appeared impossible to derive an adequate arterial input function in every patient, mainly because of the anatomic location of some tumors, the absence of a large vessel in the field of view, and strong inflow (signal saturation) effects. Therefore, a standardized arterial input function was used (23). All DCE MRI data were analyzed with software developed in-house. A pharmacokinetic 2-compartment bidirectional exchange model was used to determine K^trans (24,25). Pharmacokinetic analysis was done on a pixel-per-pixel basis. K^trans values of individual pixels were collected and tabulated in a K^trans pixel histogram. The SD was calculated, representing the heterogeneity of tumor K^trans distribution (K^trans SD) (25–27). Analysis was performed after ruling out zero values of K^trans to exclude nonperfused regions, for which the pharmacokinetic model is not valid.

Statistical Analysis

To investigate the agreement of SUV and MR_glu measurements at baseline and after 3 wk of treatment, the r² value was calculated together with the slope of the standardized values of the difference of both paired measurements. The 95% confidence interval (CI) of the slope was obtained using the adjusted bootstrap percentile method. Baseline patient characteristics were compared with ANOVA for continuous variables and with the Fisher exact and χ² tests where appropriate for qualitative variables.

The predictive value of CT-, DCE MRI–, and PET-derived measures for PFS was assessed using the percentage change from the baseline value (Δ value). PFS was defined as the time from start of treatment to the date of first documented disease progression in terms of RECIST or the date of death. Patients who had not progressed or died at the time of analysis were censored at the date of last contact. Predefined cutoff points were used and curves were compared by log rank testing. On the basis of reproducibility data, a 20% cutoff point was used for PET (i.e., values outside 1.96 × SD) (16,28). For DCE MRI, a 40% cutoff point was used for K^trans and 15% for K^trans SD (10,29). Statistical analyses were performed using SPSS (version 15.0; SPSS Inc.). Results are presented with hazard ratios (HRs), including 95% CIs and P values. A P value of less than 0.05 was considered to be significant.