Time-of-flight List-mode based motion correction for 18F-MK6240 PET imaging

1466 Purpose: 18F-MK6240 is a second-generation PET radiotracer for neurofibrillary tangle imaging, which has the potential to predict the progress of Alzheimer’s disease [Hostetler et al.2016]. 18F-MK6240 dynamic scans are prone to head motion because of long acquisition time (1-2 hrs). It is well known that head motion results in degraded image quality. In this study, we applied a time-of-flight (TOF) list-mode based motion correction approach to three human 18F-MK6240 scans and evaluated its performance by measuring both Standard Uptake Value Ratio (SUVR) and Distribution Volume Ratio (DVR). Methods: First, we divided PET list-mode data into 1-s bins and within each bin computed center of mass (COM) of the coincidences’ distribution using TOF information of events. Second, we computed the covariance matrix within a 25-s sliding window over the COM signals inside the window. The sum of the eigenvalues of the covariance matrix was used to separate the list-mode data into static and moving frames [Sun et al. 2019]. Third, we estimated motion by registering the back-projected images of these frames to a selected static frame rigidly. Finally, we corrected the motion by aligning the divided reconstructed frames with estimated motion. We applied our approach to three 120-min dynamic 18F-MK6240 scans (acquired on a GE MI Discovery PET/CT), which were identified with motion artifacts retrospectively. These scans were acquired on one Alzheimer’s disease (AD), one Mild Cognitive Impairment (MCI), and one healthy control (HC) subject. For uncorrected data, the times used to perform framing were 6×10 s, 8×15 s, 6×30 s, 8×60 s, 8×120 s, 18×300 s [Guehl et al. 2019]. For the corrected data, the times were defined according to the frame division which was resulted from the motion detection. We performed visual assessment of all the dynamic images with and without motion correction. Also, for each scan, we selected a region of interest in the cerebellum and computed SUVR and DVR with and without motion correction. For SUVR, we used a 10-min frame starting from ~ 100 min after tracer injection. For DVR, we used the Logan Graphic analysis and performed a least-squared fitting (t* = 60 min). Results: Visual assessment indicated greatly improved alignment across all the dynamic frames. Also, apparent motion artifacts in both SUVR and DVR images were reduced by motion correction. For quantification, motion correction yielded higher SUVRs (5.2%, 15.1%, and 2.8% for AD, MCI, and HC respectively, Fig. 1) and higher DVRs (9.5%, 8.2%, 2,0% for AD. MCI, and HC respectively, Fig. 2) than no-motion-correction in a cerebral cortex region at a mid-volume slice. These improvements were mainly due to more accurate time-activity-curves, resulting from the correction of intra-/inter-frame motion. Another finding was that the scans from the AD and MCI subjects seem to have more improvement than the one from the HC subject. One explanation is that the HC uptake distribution is relatively uniform across tissues and stable across frames, hence the motion has less impact in quantification; another explanation is that an HC subject tends to move less compared with an impaired subject. However, this needs to be further validated because of the limited number of scans in this study. Conclusions: In this preliminary study, we demonstrated a proposed TOF-list-mode based motion correction method. The results of patient studies indicated that it can produce improved image quality visually and quantitatively for both static and parametric 18F-MK6240 imaging.