Abstract
2311
Objectives Cardiac or respiratory gating can be used to improve resolution in NM imaging. Because gated data comes at the sacrifice of statistics, there is an interest in quantifying motion and re-combining separate gates using 3D image space morphing maps. Algorithms based on optical flow (OF) principles, like the one we are presenting, offer an approach for creating a map between a source and target image. By adding sub-iteration updates to a common OF approach, we developed an algorithm that will not be ruined by noise and where the motion vectors are calculated independent of activity concentration - thus it is robust over the whole body and for ranging conditions.
Methods The algorithm works by determining the direction of optical flow, on a voxel by voxel basis, and interpolating displacement parameters at subvoxel steps over many iterations. The displacement vectors continue to grow until a minimal difference between source and target voxel values are reached. The principle can be likened to dropped water organizing itself into puddles (through gravity). To minimize noise being classified as motion, smoothing, truncation, and gate combination strategies were also employed. To test the accuracy of the algorithm, motion maps were generated to map source gate distributions onto target gates in simulated phantoms and a collection of 78 μPET rat scans.
Results By applying our algorithm, we were able to robustly characterize 3D motion maps for simulations and gated μPET scans. Processing each scan volume required ~ 1min. The absolute sum difference between pairs of source and target gated μPET images was reduced by an average of 78% (SD 8%), through the application of the motion map, indicating the greatly increased similarity in the corrected images.
Conclusions We present a new fully automated algorithm to quantify extent of motion in potentially noisy NM images, which can be used to combine statistics and visualize anatomical motion. We are exploring if modifications can be made to ensure mass conservation and minimize the influence of noise.
Research Support USIEF Fulbrigh