Anatomical priors to improve image quality in small-animal SPECT/CT

With the advent of SPECT/CT, high-quality coregistered anatomical information is now readily available and can be incorporated into the SPECT reconstruction. Previous work suggests that anatomical priors may be useful in regularizing iterative reconstruction. This study used computer simulations to investigate the impact of incorporating anatomical priors into a rescaled block iterative(RBI)-maximum a posteriori(MAP) reconstruction of small-animal pinhole SPECT data, particularly in those regions where there was no anatomical boundary information. Two computer phantoms were modeled: 1) a hot-rod phantom that included one rod in which there was an activity boundary but no corresponding anatomical boundary and 2) sestamibi uptake in a MOBY mouse phantom for which the anatomical prior did not differentiate between myocardial chambers and walls. Three reconstruction algorithms were considered: RBI with post-reconstruction Gaussian filtering, RBI-MAP with a Gibbs prior, and RBI-MAP with a Gibbs prior restricted to regions defined from anatomical boundaries. The Gaussian filter width and Gibbs prior parameters were optimized to minimize the absolute difference from the true activity distribution. For both phantoms, the mean absolute error over the entire image was improved using the MAP algorithms compared to RBI. However, the MAP algorithm also led to resolution loss in the image. The use of anatomical priors further reduced the mean absolute error and greatly reduced the loss of resolution at anatomically defined boundaries. Anatomical priors also led to a small reduction in absolute error even in the region of activity change that had no corresponding anatomical boundary. For the hot-rod phantom, the mean absolute error was reduced by 60% through the incorporation of anatomical priors with a RBI-MAP reconstruction compared to RBI alone. For the MOBY phantom simulation, the error in the heart region was reduced by 11% and the contrast between the myocardial walls and ventricle was improved by 28%. Use of anatomical priors in small-animal SPECT/CT imaging is promising and merits additional investigation.