Abstract
1324
Objectives We aimed to develop an automated method to minimize spatial blurring and artifacts caused by patient motion in high resolution dynamic cardiac PET performed with CT attenuation correction.
Methods We considered 15 consecutive stress/rest datasets (30 datasets, 11 male and 4 female) acquired on a Siemens Biograph-64 scanner in list mode with a new 18F based agent (BMS747158). For motion correction, 15-min stress and rest images were reconstructed into 10 dynamic frames (90 sec each) and voxel size of 2.6x2.6x2 (mm). Adenosine stress was performed in 6 and exercise stress in 9 patients. After median filtering, automated frame-to-frame rigid registration is performed on most prominent feature (myocardium) with normalized cross-correlation and gradient descent minimization, in which sample voxels are randomly selected for computational efficiency.
Results The registration technique had 100% success rate in removing left ventricular motion for stress and rest images, as visually assessed by dynamic cine display. The average registration time was 94±26 sec. The maximum correction for stress and rest were (2.91, 2.11, 6.21) mm and (2.4, 1.5, 5.87) mm, for x-, y- and z- axes, respectively. The mean translational motion corrections for stress and rest were (0.86±0.71, 0.54±0.48, 3.26) mm and (0.76±0.56, 0.59±0.45, 1.95±1.44) mm for x-, y- and z- axes, respectively. Motion correction in 5 stress subjects was greater than 2 voxels (≥4 mm) and in 11 stress subjects was greater than 1 voxel (≥ 2 mm). At rest, 1 subject had motion greater than 2 voxels and 4 greater than 1 voxel.
Conclusions Motion during cardiac PET occurs frequently in the up/down (z) direction and is often more than 2 pixels on stress scans. Fully automatic registration of dynamic 90 sec PET frames can be performed accurately, potentially improving accuracy of high resolution cardiac PET.
Research Support NIH/NHLBI 5R01HL089765-0