Sixteen patients with esophageal and gastroesophageal junctional (GEJ) tumors were included in the study. Patient characteristics are described in detail in another study (Linder et al. 2019); in brief, inclusion criteria were potentially resectable tumors (T1-4a, N1-3, M0) and histologically verified esophageal or GEJ cancer (Siewert I and II). The patients were asked for study participation in the clinical setting after cancer diagnosis. PET-MRI was performed in conjunction with the routine PET-CT investigation to conclude clinical staging. All patients provided written informed consent to participate in the study, and approval was granted from the regional medical ethics committee in Uppsala (DNR 2014/551).
Image acquisition and reconstruction
All patients underwent a 6-min PET scan on a 3T Signa PET-MRI system (GE Healthcare) 109 ± 21 min after injection of 4 MBq/kg 18F-FDG. The PET system has an axial and transaxial field of view (FOV) of 25 cm and 60 cm, respectively, producing 89 image planes with a slice thickness of 2.8 mm. Data were acquired in list mode to enable reconstruction of data in different time frames and breathing gates retrospectively.
External hardware gating was performed using an MRI-compatible respiratory bellow device, and a quiescent phase-based gating method (Q.Static, GE Healthcare) was used for the whole 6-min scan. Static gates were automatically extracted during end-expiration of the breathing cycle to form a gating phase free from motion (Soussan et al. 2016). The external gating triggers were stored in the PET list file, which enables retrospective unlisting of gated data during the quiescent phase. The quiescent phase includes 50% of the total breathing cycle, resulting in maintained PET coincidence data equivalent to a 3-min static scan.
The DDG was performed using MotionFree (GE Healthcare), which uses PCA to derive the respiratory waveforms directly from the PET coincidence data. In short, data are binned into 0.5-s sinograms, and the frequency of the motion can be determined by using the Fourier transform. Respiratory motion is defined as motion originating from frequencies in the range of 0.1–0.4 Hz, implying a respiratory cycle of 2.5–10 s. To establish the impact of the motion on the data, an R value is calculated as the ratio between the peak value within this respiratory frequency range and the mean value above the respiratory frequency range.
The whole 6-min PET list file was transferred to a Matlab-based toolbox containing the DDG MotionFree package (Duetto v02.03, GE Healthcare). In the toolbox, respiratory waveforms are derived from the PET-coincidence data, and DDG respiratory gating triggers are stored in the PET list file in the same way as for the external gating triggers. For the data-driven gating, an R value was given for each patient, describing the impact of respiratory motion on the data, where a threshold of R=15 is the default value recommendation for MotionFree.
Raw files were created in the toolbox by unlisting the static gates from the whole 6-min scan list file. For comparison with the gated data and for investigation of normal image quality variation in a scan-rescan situation, two 3-min static raw files (0–3 min and 3–6 min) were also created for each patient.
All images were reconstructed using TF-OSEM with resolution recovery, 2 iterations, 28 subsets, and a 3-mm Gaussian post-processing filter. A 60-cm FOV was used with a 192 × 192 reconstruction matrix, resulting in a 3.125 × 3.125 × 2.80 mm3 voxel size.
SUV images were calculated by dividing the activity concentrations with the amount of injected activity per body weight. Images were visually assessed with respect to image artifacts and reconstruction errors. Esophageal lesions and lymph nodes were identified on the PET images, and SUVmax as well as lesion volume, based on a 41% SUVmax threshold (Boellaard et al. 2015), were measured using Hermes Affinity Viewer 1.1 (Hermes Medical Solutions). For image noise measurement, a spherical volume of interest with a diameter of 3 cm was placed in the liver in all images, and SUV mean values together with standard deviations were measured. Noise level was defined as the standard deviation divided by the mean SUV. To estimate the effect of lesion position on the efficacy of DDG, the vertical distance from the center of the lesion to the top of the liver was measured. Lesions below 1 cm3 were represented separately in graphs.
Results are presented as mean ± standard deviation unless otherwise specified. Data representing SUVmax and lesion volume were not normally distributed, while data representing changes in these parameters were normally distributed according to Shapiro-Wilk normality tests. To determine whether SUVmax and lesion volume measured in images with either of the two respiratory motion correction methods, external hardware gating and DDG, differed from those of a static non motion-corrected image, a non-parametric paired Wilcoxon test was used. Correlation between both gating methods and non-gated images was assessed using linear regression and Spearman correlation coefficients for SUVmax and lesion volume. To investigate agreement between DDG and external hardware gating, Bland-Altman analysis was performed for SUVmax and lesion volume values. Correlation between the distance from the lesion to the diaphragm, and the increase in SUVmax respectively the decrease in lesion volume, were assessed using Pearson correlation coefficients. Similarly, the correlation between the DDG R value, and the increase in SUVmax respectively the decrease in lesion volume, were assessed using Pearson correlation coefficients.