Abstract
10
Objectives: Respiratory motion blur in static whole body PET (SWB) imaging is unavoidable, leading to underestimation of radiotracer concentration, and decreased lesion detectability. Traditional methods of motion correction such as phase based gating(PBG) and amplitude based optimal gating (ABOG) reduce motion blur but increase image noise due to use of a fraction of the data. Elastic motion correction (EMC) techniques in PET overcome this challenge by using all the acquired data. One method reconstructs PGB images, registers the PET images to a target image, and then averages the registered images (RRA). A more recent method registers ABOG images to SWB images with a blurring kernel using mass preserved optical flow. This calculated blurring kernel is used in an iterative reconstruction which performs elastic motion correction via deblurring(EMCDB) and uses all of the data during the reconstruction. In this work the impact on image quality and quantification of the RRA and EMCDB motion correction techniques is evaluated with patient studies and a phantom study.
Methods: 46 patients were included in this study. In addition, a phantom was constructed to drive 5 spheres of inner diameters 10 mm, 13mm, 17mm, 22 mm, and 28 mm. The phantom was designed so that the spheres moved on non-parallel trajectories, to simulate lung tumor motion. Four scans of the phantom were performed, one with no sphere motion, and three scans driven by a patient respiratory waveform with motion amplitudes of 1, 2, and 3 cm. The phantom background and sphere activity concentration were 0.1 and 0.5 µCi/cc F-18 respectively. All scans were acquired on a 4 ring Siemens mCT with continuous bed motion. A slower table speed of 0.5 mm/s was prescribed over a 30 cm region covering the lesions impacted by motion and was centered over the spheres for the phantom scan. The respiratory waveform was acquired with the Anzai belt. For each scan, three reconstructions were performed: SWB, EMCDB, and 6 gates PBG for RRA. The RRA was performed using phase 4/6 for the target phase. For the EMCDB reconstruction, a 35% ABOG reference image was used. All reconstructions were performed with 2 iterations, 21 subsets, TOF, PSF, and a 5mm Gaussian filter. Measurements of SUVmax, SUVpeak of 65 lesions, and the phantom spheres were made. To assess image quality, the STDev of healthy lung and liver was measured in all patients, as well in the phantom background. For the patient scans the percent change of the average values were made with respect to the SWB reconstruction. For the phantom scans, the gold standard was the scan with no motion. The recovery coefficient of SUVmax (RCmax) and SUVpeak (RCpeak) was calculated for each phantom scan for all spheres. For the phantom, the percent change in image noise was calculated in comparison to the SWB reconstructions.
Results: For the 65 tumors measured, SUVmax (SUVpeak) increased by 8.9% (9.7%) for RRA and 19.3% (11.1%) for EMCDB. The increases in background noise in the lung and liver were higher for RRA(19.6% & 66.1%) compared to EMCDB(3.3% & 14.8%). For the phantom evaluation, the average RCmax for all spheres were similar for RRA and EMCDB and improved in comparison to SWB across all amplitudes: RCmax for (SWB / RRA / EMCDB) 1cm amplitude (0.95 / 0.95 / 1.04), 2cm amplitude (0.81 / 0.97 / 0.96), 3cm amplitude (0.66 / 0.92 / 0.93). The increase in RCpeak for (SWB / RRA / EMCDB) 1cm amplitude (0.97 / 0.97 / 1.02), 2cm amplitude (0.85 / 1.03 / 0.94), 3cm amplitude (0.71 / 0.96 / 0.91). Similar to the patient scans, the average increase in background noise was higher for RRA(57%) than for EMCDB(19%). Conclusion: Both methods of respiratory elastic motion correction reduce motion blur, increase measured SUV in patient images, and improve the activity recovery in the phantom studies. Although both methods use all of the acquired PET data, EMCDB shows an advantage by introducing less noise into the motion corrected images in comparison to the RRA methodology.