Temporal compression for dynamic positron emission tomography via principal component analysis in the sinogram domain

We compare dynamic PET temporal compression using optimal sampling schedule design, principal component analysis (PCA) in the image domain, and principal component analysis in the sinogram domain. For region-of-interest quantification, sinogram-domain PCA is combined with the Huesman algorithm to quantify from the sinograms directly without requiring full frame reconstruction. Using a simulated phantom FDG brain study and three clinical studies, we evaluate the fidelity of the compressed data for estimation of local cerebral metabolic rate of glucose by a four-compartment model. Our results show that using a (noise-normalized) PCA in the sinogram-domain gives similar compression ratio and quantitative accuracy to OSS, but with much better precision. These results indicate that sinogram-domain PCA for dynamic PET can be a useful preprocessing stage for PET compression and estimation applications.