Abstract
1638
Objectives: HighlY constrained back PRojection (HYPR) was initially introduced for reconstruction of highly undersampled MRI time series data. The general HYPR approach involves the formation of a composite image from many or all of the images in a time series to take advantage of the high SNR summed data. HYPR weighting images are formed by a ratio of convolved versions of the time sequence and the composite image, thus feeding most of the SNR accumulated in the entire scan into each of the time frames. Preliminary research was conducted on human and small animal PET data to assess the utility of HYPR processing.
Methods: For qualitative comparison, HYPR processing was performed on human PET scans of [O-15]water studies with short frame duration (2 seconds) and reconstructed images dominated by noise. To assess the quantitative accuracy, phantom studies were acquired on a small animal PET scanner filled with 18F solution and air- and 11C-filled inserts.
Results: There were dramatic improvements in image quality for the human scans of [O-15]water (see figure), with the HYPR image providing clear visualization of the brain which was imperceptible in the original image. For quantitative evaluation of the phantom study, voxel based analysis of measured concentration yielded accurate activity recovery with a 3-fold reduction in voxel standard error in all regions. Within the 11C region, the measured half life was consistent with the original data with no significant bias evident in the HYPR data.
Conclusions: HYPR processing holds great potential in nuclear medicine imaging for all applications suffering from low SNR in dynamic scans, including the generation of voxel-based parametric images, coregistration of image frames and visualization of rapid radiotracer uptake and distribution.
- Society of Nuclear Medicine, Inc.