Cardiac 15O-water PET motion correction using patient specific synthetic dynamic images

Patient motion is a common problem in cardiac PET, impairing the diagnostic accuracy of hemodynamically significant CAD for both qualitative and quantitative assessments. Motion correction in cardiac PET is challenging in general because of the time-varying distribution of the tracer, and even more so for 15O-water due to its lack of retention in the myocardium. The aim of the present work was to evaluate a frame-by-frame motion correction method using a patient specific synthetic dynamic reference.

Motion correction was implemented as follows. The subject’s CT scan was segmented using Total Segmentator (Wasserthal J et al.) and resulting volumes of interest were then projected over the original dynamic PET scans to obtain time activity cures (TACs) for each segmented structure. A patient specific non-moving synthetic dynamic reference was created based on the segmented CT and extracted TACs. Then, each individual frame of the dynamic PET scan was aligned to the corresponding synthetic frame using rigid co-registration (Figure 1). To test the accuracy of the method, 10 cardiac 15O-water stress scans from patients with known or suspected CAD were included. The scans were visually inspected and determined to be without motion before four different types of motion were added as previously described (Nordstrom et al.), and then corrected using the proposed method. All scans were then analyzed using aQuant software and MBF values from scans with and without motion correction were compared to MBF values from the original scans without motion.

After motion correction, average bias in MBF for all motions decreased from 22.1% to 9.8% in the LAD region, from 16.8% to 12.0% in the RCA region, and from 8.4% to 6.2% in the LCX region. The effect of motion correction in the LAD region for each separate motion is seen in Figure 2.

Motion correction clearly reduces the impact of motion on quantitative MBF, especially for motion caused by cough, but motion artefacts are not completely eliminated. The proposed method addresses the challenges associated with frame-by-frame motion correction due to time varying tracer distributionFigure 1.

An example of motion correction for an added cough simulated as a 2 cm anterior displacement during one frame at peak first pass. Top row shows the synthetic reference, second row shows the original scan with an added cough in frame 6 and the third row shows the motion corrected scan. The bottom row shows the magnitude of motion correction in x, y and z direction.