TY - JOUR T1 - Data-driven motion correction for cardiac PET<strong/> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 375 LP - 375 VL - 61 IS - supplement 1 AU - Ian Armstrong AU - Charles Hayden AU - Parthiban Arumugam Y1 - 2020/05/01 UR - http://jnm.snmjournals.org/content/61/supplement_1/375.abstract N2 - 375Background: The increasing sensitivity and spatial resolution of PET-CT scanners exposes further the degrading influence of internal organ motion on image quality. This is particularly relevant in myocardial perfusion PET where, in additional to periodic respiratory motion, the effect of vasodilator stressing agents have been shown to produce severe motion artefacts in patients. We developed a prototype data-driven motion correction (DDMC) algorithm specifically for cardiac PET. The algorithm tracks the position of the heart in histo-projection space with a temporal resolution of 1 second. To remove this motion, the heart is shifted axially and aligned to a reference position. Unlike current respiratory gating techniques, the algorithm assumes no periodicity to the heart’s displacement and will deal with both periodic and non-periodic motion of the heart. Methods: A cardiac torso phantom, with added cardiac defect, was filled with F-18 and scanned for 3 minutes in list mode on a Biograph Vision 600 (Siemens Medical Solutions Inc., Knoxville, TN, United States). Scans were performed with the phantom stationary and with random back and forth motion over an approximate range of 65 mm in the axial direction during the scan. DDMC was applied to the data with motion. Three image sets were produced: stationary, motion without correction and motion with DDMC. Axial profiles were drawn through the walls of the cardiac insert and FWHM measured. The Total Perfusion Deficit (TPD) was calculated using Cedars Sinai QPET (Cedars Sinai Medical Centre, Los Angeles, CA, United States). 46 images (17 rest, 29 stress) from 36 patients (28M; 8F) who underwent clinically indicated rubidium-82 scans were included based upon a visual assessment of suspected motion blurring. 20 mCi rubidium-82 was administered during rest and stress and with a scan time of 5 minutes. For the 29 stress studies, 24 were adenosine stress and 5 with regadenoson. DDMC algorithm was applied to the data and non-corrected and corrected images were reconstructed using PSF+TOF reconstruction with 4 iterations, 5 subsets and 6.0 mm FWHM Gaussian filter, isotropic 1.6 mm voxels and a framing of 120 to 300 second after the acquisition start. Images were blinded and was scored by an experienced physician in cardiac PET reporting for quality (non-diagnostic, adequate and good) and perceived motion (none, mild, moderate, severe). Results: The FWHM of the cardiac insert axial profiles was 11.7 mm and 13.2 mm for the stationary and moving phantom with DDMC respectively. The TPD was 9% for the stationary phantom, and 10% and 46% for moving phantom with and without DDMC respectively. For the 46 rubidium images, quality was rated as follows: non-corrected images: 10 non-diagnostic, 31 adequate and 5 good; DDMC images: 0 non-diagnostic, 6 adequate and 40 good. Perceived motion was rated as follows: non-corrected images: 4 no motion, 24 mild, 11 moderate and 7 severe; DDMC images: 42 no motion and 4 mild. Conclusions: From phantom data, we have demonstrated that application of the data-driven motion correction algorithm has the potential to completely correct varying degrees of cardiac motion. For clinical cases, the DDMC algorithm was able to restore all non-diagnostic cases and allow clinical interpretation. As motion degraded images are usually non diagnostic this method, once clinically validated, will avoid the need for a repeat scan and further radiation exposure. ER -