Abstract
1753
Objectives SUVR is usually used for amyloid imaging, but it required an assumption of equilibrium of radioactivity between tissues and blood. If a kinetic analysis is conducted, more reliable BPND images can be derived [1]. And, the early images in a dynamic scan is expected to be substitutes for FDG [2,3]. The dynamic scan is therefore useful for a clinical amyloid imaging. However, noise in a voxel-based tissue time activity (tTAC) is problematic. A kinetic-based clustering algorithm named CAKS was proposed [4] that address the bad noise statistics in tTAC. This study investigates the possibility of CAKS to BPND amyloid imaging.
Methods CAKS clusters voxels based on a rate constant of k2 assuming the kinetics obeying 1-tissue-2-compartment model. A ratio between tTAC and tTAC multiplied by acquisition time is a function of k2, and such voxels having the similar ratio are clustered. The tTACs are then averaged to improve the noise statistics. BPND is estimated using Logan graphical algorithm (LGA). In CAKS, k2 is common for a voxels belonging to the same cluster, but K1 and a BPND is estimated to the every voxels because an amplitude of tTAC is proportional to K1. A simulation study was conducted in which a set of tTACs having clinically feasible rate constants and noise was generated, in which BPND varied from 2 to 4. And the algorithm was applied to PiB dynamic data.
Results The bias of estimated VT was less than 2%, and it tends to increase in the larger VT. The typical clinical BPND images are presented in the case of Alzheimer dementia. (A) is the image with CAKS. LGA underestimate BPND in the presence of noise in tTAC, but CAKS reduced the noise, and higher BPND images can be derived. And the images with CAKS were smoother than those without CAKS (B).
Conclusions The assumption of CAKS to the kinetics does not affect the performance. We conclude that CAKS is a potential algorithm for amyloid imaging.