Performance assessment of motion correction for different distributions and count levels

Motion correction methods are used to compensate for subject motion during a positron emission tomography (PET) scan, reducing the degradation in image quality and quantification and consequently improving the accuracy of data analysis. In this work a frame-by-frame realignment method applied to a [¹¹C]raclopride dynamic brain scan has been assessed, where the scan has 26 frames of different count levels and a varying spatiotemporal distribution. The performance of the motion correction in its two modes (normal and enhanced) is assessed using list-mode bootstrapping giving insights into the precision (and not accuracy) of the motion correction with respect to both the count level and distribution of the radioligand. It is found that the precision of motion correction deteriorates with decreasing count level (as expected in short low-count frames) and varies according to the radial and axial position in the field of view (FOV). Also the precision is affected by the varying distribution of the radioligand in frames with similar count levels, i.e., the greater concentration of the radioactivity in the striatal regions (and lower radioactivity concentration in the cortical areas) decreases the precision. Of particular note is the observation that the precision is also considerably degraded when the enhanced mode is used-indicating that the improved motion correction accuracy comes at the expense of increased variability (hence the enhanced mode simply offers a different bias-variance trade-off). It becomes clear that although such frame-by-frame motion correction method does not require an external motion tracking device it is consequently affected by the count level in each frame, radioligand distribution, voxel position in the FOV and the mode of coregistration. Such assessment can be applied to any coregistration of either PET or SPECT images especially with low counts.