Ziel/Aim:
PET is well known to be affected by respiration. Therefore motion correction methods
have been introduced using optical flow techniques. In this study, optical flow parameters
were optimized based on PET to receive optimal motion vector fields covering the real
physiological respiratory motion.
Methodik/Methods:
PET/CT listmode data and a respiratory signal (ANZAI belt system) of 17 patients were
acquired on an integrated PET/CT, 1h p.i. of F-18-FDG (4MBq/kg body weight). Data
was sorted into 10 respiratory gates with the ANZAI signal and reconstructed. Optical
flow, based on a multi-level approach by Horn-Schunck (1), was applied to gate 10
(reference gate 1) by varying the intrinsic parameters (480 combinations): regularization
α, step size of the algorithm (tau), number of iterations, number of multi-levels
and mode. The evaluation of these vector fields was based on a volume of 11 × 11 ×
11 voxels with a FDG positive lesion at the center. For each lesion, the mean 3D vector
length was calculated and compared to the manually determined displacement. In addition,
the angle between the computed and manually determined vector was calculated. The
parameter set leading to the lowest deviation regarding vector length and orientation
is assumed to best cover the real physiological motion.
Ergebnisse/Results:
40 lesions were included in the optimization process. Considering all evaluated cases,
different parameter sets exist leading to similar optimal results regarding vector
length and orientation. We found one common parameter set for all lesions with parameters
alpha = 0.01, tau = 0.1, number of iterations = 7500, number of multi-level = 10,
mode = no mass preservation.
Schlussfolgerungen/Conclusions:
Parameter optimization for optical flow is an important preprocessing step in motion
detection approaches to achieve reasonable physiological motion vectors. A common
optimal set of parameters could be defined that results in excellent motion vector
fields for all evaluated PET cases.
Literatur/References:
[1] M. Dawood, et al., “Respiratory motion correction in 3-D PET data with advanced
optical flow algorithms.” IEEE transactions on medical imaging, vol. 27, no. 8, pp.
1164 – 75, 2008.
