Quantitative Analysis of Response to Treatment with Erlotinib in Advanced Non–Small Cell Lung Cancer Using ¹⁸F-FDG and 3′-Deoxy-3′-¹⁸F-Fluorothymidine PET

Patients

Between September 2007 and September 2009, patients with advanced-stage non–small cell lung cancer were recruited into the trial at the University Hospital of Cologne (8). Inclusion criteria were at least 1 measurable target lesion, age over 18 y, Eastern Cooperative Oncology Group performance status of 0–2, serum creatinine less than 1.7 mg/dL, no decompensated liver failure, no sign of hyperthyroidism, normal blood glucose levels, and no prior systemic treatment. Written informed consent was obtained from all patients, and the trial was approved by the local ethics committee, the responsible Federal Institute for Drugs and Medical Devices (BfArM), the responsible federal state authorities of North Rhine-Westphalia, and the Federal Office for Radiation Protection (BfS) (clinical trials, NCT00568841). To investigate the study objectives mentioned, previously published datasets were included in the current analysis (8).

Treatment

Erlotinib was administered as first-line therapy at a dose of 150 mg once daily for 6 wk or until documented progression. Doses were reduced in accordance with the manufacturer's recommendations in the case of adverse events, and treatment was discontinued on development of grade IV toxicity. Two patients temporarily interrupted treatment for more than 2 wk because of adverse reactions. During the treatment pause, the tumor recurred, but new response was observed in 1 patient when treatment was continued at a reduced dose. One patient underwent additional radiotherapy of the target lesion.

Image Acquisition

¹⁸F-FDG PET and ¹⁸F-FLT PET scans were performed within 10 d before therapy and at 1 and 6 wk after the start of erlotinib therapy. ¹⁸F-FLT and ¹⁸F-FDG were synthesized as previously described (14,15). PET images were obtained using an ECAT EXACT 47 (Siemens) after the patient had fasted for 6 h. The time between tracer injection and acquisition of data was 59 ± 14 and 58 ± 15 min for ¹⁸F-FDG and ¹⁸F-FLT, respectively. On average, 365 ± 30 MBq of ¹⁸F-FDG and 305 ± 89 MBq of ¹⁸F-FLT were injected. The minimal time between time from injection to imaging was 36 min, and the maximal time was 96 min. For a single patient, scanning differed by less than 15 min in all cases. The attenuation-corrected scan trajectory covered 90 cm (6 bed positions: 5-min emission, 3-min transmission). All scans were corrected for decay, dead time, scatter, and randoms and reconstructed by ordered-subset expectation maximization. The same scanner, same protocol for acquisition, and same software for reconstruction were used for all patients analyzed (ordered-subset expectation maximization using 4 iterations and 16 subsets; zoom, 1; image matrix size, 128 × 128; and gaussian postsmoothing of 5 mm in full width at half maximum). The final image spatial resolution was approximately 7 mm in full width at half maximum near the center of the field of view, using these settings.

Image Analysis

As proposed by the PET Response Criteria in Solid Tumors (PERCIST) 1.0 guideline (16), measurements were performed in the hottest single lesion and additionally in the sum of up to 5 measurable target lesions. These were analyzed with no more than 2 lesions per organ, selecting the lesions with the highest maximum SUV (SUV_max). The target lesions in ¹⁸F-FDG PET and ¹⁸F-FLT PET did not have to be identical, but wherever possible, lesions were selected that were suitable for analysis with both tracers and for Response Evaluation Criteria in Solid Tumors (RECIST) (17) measurements. In accordance with the PERCIST 1.0 guideline (16), the 2-dimensional (2D) peak SUV (SUV_2Dpeak) of the hottest baseline single tumor lesion in ¹⁸F-FDG PET was at least 1.5-fold greater than the mean baseline liver SUV plus 2 SDs in all patients. In contrast, for ¹⁸F-FLT PET analyses the use of liver as background control was not possible, because, in general, ¹⁸F-FLT SUVs in liver were higher than SUVs in tumor lesions (12,18).

The tumor volume of interest (VOI) for SUV calculation was defined using a 3-dimensional (3D) region-growing algorithm, implemented with software developed in-house at the VU University Medical Center (19,20) that makes use of the 3D search algorithm in the IDL software package (version 6.2; Research Systems Inc.). First, SUV_max normalized to body weight was determined for each lesion, using the voxel with the maximum uptake on reconstructed PET images without additional rebinning, resampling, or smoothing. The SUV_2Dpeak was determined using a 1.2-cm-diameter, fixed-size circular region of interest defined in the axial plane, automatically centered on the tumor area with maximum uptake. The 3D SUV peak (SUV_3Dpeak) was determined using a 1.2-cm-diameter spheric VOI automatically centered on the tumor area with maximum uptake. In contrast to previously published data (8), 2D and 3D peak VOIs were defined fully automatically without any manual correction. In addition, the following SUVs were obtained: the 3D isocontour at 41% of the maximum pixel value adapted for background (SUV_A41), 3D isocontour at 50% of the maximum pixel value (SUV₅₀), 3D isocontour at 50% adapted for background (SUV_A50), 3D isocontour at 70% of the maximum pixel value (SUV₇₀), and 3D isocontour at 70% adapted for background (SUV_A70). The background-adapted VOI was generated by first generating a 3D VOI using the threshold of 70% of maximum pixel value. This first VOI roughly indicates the boundary of the metabolic volume of the primary tumor. Next, voxels located 2 cm from this initially estimated tumor boundary are used to estimate the average local background value. However, voxels with SUVs greater than 3 are excluded to avoid inclusion of tumor voxels in the background estimation. A background-adapted threshold value is then obtained by taking, for example, 50% of the maximum tumor pixel value plus the average background value. For example, this means that the final SUV_A50 threshold will be 1.4 if background equals 1 and tumor SUV_max equals 1.8. Higher background values will thus automatically shift the threshold to higher values. This adapted background threshold value is then used in a final region-growing step, using the location of the maximum tumor pixel value as a starting point, to generate the final VOI (21). The relative SUV threshold level (SUV_RTL) was calculated where possible, as described elsewhere (22). The metabolic tumor volume was also estimated on the basis of the SUV₅₀ isocontour (VOL₅₀). The product of VOL₅₀ and SUV₅₀ was calculated for both ¹⁸F-FDG and ¹⁸F-FLT scans and named total lesion glycolysis (TLG) (23) and total lesion proliferation (TLP). Data for the SUV_2Dpeak, using a 1.2-cm-diameter, fixed-size circle manually set on the tumor area with the highest uptake and the SUV_max derived by commercially available software, have already been shown (8).

Response Assessment Based on PET

Metabolic response was defined according to the PERCIST guidelines (16): a reduction in ¹⁸F-FDG or ¹⁸F-FLT uptake of 30% or greater was defined as a metabolic response for all SUVs; for VOL₅₀, TLG, and TLP, a metabolic response was defined as a reduction of uptake of 45% or greater for both ¹⁸F-FDG and ¹⁸F-FLT. Receiver-operating-characteristic analyses were performed to make sure that cutoff values applied were acceptable for determination of metabolic response. For each patient and each scan, the SUV of the lesion with the highest SUV_max at baseline was compared with the SUV of the lesion with the highest SUV_max on the later PET scans performed after 1 and 6 wk (single hottest SUV). The determination of the single hottest lesion was independent of the previous scan. Additionally, for each patient and every scan, up to 5 lesions with the highest SUV_max were summed for all identified lesions (sum SUV) at baseline and compared with the sum SUV on the later PET scans after 1 and 6 wk.

Statistical Analysis

The objective was to compare different SUVs and metabolic tumor volume measurement to assess their ability to predict progression-free survival (PFS) under ongoing therapy with erlotinib. PFS was defined as the time between start of medication and death or progression under ongoing medication. The analysis of PFS was performed using Kaplan–Meier estimates and log-rank tests with SPSS statistical software (version 17.0; SPSS).