Optimization of Optical Flow Parameters for Whole-body PET Motion Correction

Abstract

The quality of whole-body PET (18F-FDG) images is often degraded by respiratory motion due to long scan times. Motion correction methods have been proposed. Some of these methods use optical flow techniques (OF) to estimate and correct motion. As the estimated motion vectors have a high impact on the result of the motion corrected reconstruction, it is important to provide motion vector fields that cover the true physiological motion. Therefore, an optimization procedure of OF parameters is proposed. For the optimization, 40 lesions, detected in whole-body PET data of 17 patients were analyzed. The used OF implementation is based on a multi-level approach by Horn-Schunck. For all lesions, the motion was manually estimated between the first and tenth respiratory gate. Afterwards, OF was applied to gate 10 (reference gate 1) by varying the following intrinsic parameters (480 combinations): regularization, step size of the algorithm, iterations, multi-level and mode. For each lesion, the mean 3D vector length and mean angle were calculated and compared to the manually determined ones. Furthermore, a difference image was calculated between the reference image and the respective motion corrected image. Based on the study it was possible to determine a common parameter set for all lesions, which can be included in the subsequent motion correction process. After the optimization the motion vector field calculation was performed with the optimized parameters on whole-body PET images within the motion correction approach. This showed comparable motion vector field results at the tumor region with vector-stabilizing influences of the surrounding areas. In conclusion the OF parameter optimization is an important preprocessing step in motion correction approaches to achieve best image results taking into account reasonable physiological motion.