Partial volume compensation for spill-in and spill-out in cardiac PET imaging



Yong Du, Igal Madar, Martin J. Stumpf, and Eric C. Frey





In PET imaging finite spatial resolution results in partial volume effects including spill-out and spill-in of activity to and from neighboring regions. In cardiac imaging, it is important to compensate for both effects when neighboring regions have high activity concentrations, e.g., the liver or, at early time points in dynamic studies, the cardiac blood pool. We have modified a brain PET partial volume compensation (PVC) method that considers both effects and applied to the cardiac PET imaging.



In this method the images were segmented into 3 regional images, the left ventricular (LV) wall, the LV blood pool (BP) and the rest of the image based on region-of-interest (ROI) maps (voxel values 1 inside the ROI, 0 outside) obtained from high resolution contrast CT images. The spill-in for each regional image was estimated by blurring the images of the other two ROIs then subtracting them from the region under consideration. Here "blurring" refers to a convolution with a Gaussian shaped point spread function of the PET system. To compensate for the spill-out from the LV wall or BP, a template was created for the region by blurring the uniform ROI map image for the region and then dividing the spill-in compensated regional image by the template image. The procedures were applied iteratively as follows: compensate for spill-in and spill-out in LV wall; compensate for spill-in and spill-out in BP; and, finally, compensate for spill-in for the remainder. We applied this PVC method to the dynamic PET studies of dog cardiac perfusion imaging with agent F-18-fluorobnzyl triphenyl phosphonium. Dynamic scans of increasing intervals were acquired for 60 min, during which samples of arterial blood were collected. Blood activity was counted to generate tracer time activity profile in the blood pool. We compared the activity ratios of the ischemic region to non-ischemic regions. The time activity curve of the LV wall uptake and BP clearance were also plotted and compared with direct blood counting.



The compensated image has more uniform uptake in both the LV wall and BP. The PVC removed the spill-in from LV wall to the BP, especially when the myocardial wall to blood activity ratio was very high. The BP time activity curve was more close to the true blood counting after PVC. The LV wall activity increased after PVC, indicating of a reducing of spill-out effects.



We have developed a PVC method that compensates for both spill-in and spill-out in cardiac imaging. Application of the method to dynamic PET studies showed great reduction of spill-in in BP time activity curve. This could potentially affect the parameters estimation when the BP activity is used as the input function.









Time activity curve of LV blood pool clearance and LV wall uptake. PVC greatly reduced the spill-in from high activity in the LV wall into the blood pool. The PVC results are closer to that of direct blood counting. Thus using compensated blood pool as input function for dynamic analysis could potentially greatly improve the accuracy compared to non-PVC data. In figure on the right we can see PVC also increased the activity estimation of LV wall by reducing the effects of spill-out. Here ¡°is¡± means ischemic area and ¡°nis¡± means non-ischemic region.




Transaxial images of LV wall before and after PVC. One slice of high-resolution CT images with LV wall mask is also shown. After PVC the image looks more uniform.


Normalized circumferential profiles of above transaxial images before and after PVC for 30-50 minute image. The activity ratios between ischemic and non-ischemic regions were essentially unchanged by PVC. This is reasonable because the myocardial wall has a much higher activity compared to surround regions and thus the effect of spill-in from other regions is relatively small. Effect of spill-out on the activity ratios is small but do have a big impact on the absolute measurement as shown in the LV wall time-activity curves depicted in the previous figure.