Chapter 22 - Sinogram Recovery of Dynamic PET Using Principal Component Analysis and Projections onto Convex Sets

Abstract

Positron emission tomography (PET) is typically limited by poor spatial resolution due to the finite size of the detector elements and Poisson noise due to nuclear count statistics. Noise becomes particularly limiting in dynamic studies with low count rates per individual frame. Unfortunately, most resolution enhancement schemes amplify high-frequency noise. Furthermore, tomographic reconstruction using simple filtered back projection requires filtering out high-frequency information to prevent further amplification of noise. Thus, a trade-off between spatial resolution and noise is usually made in the reconstruction. Principal component analysis (PCA) has been previously used to extract the relevant temporal kinetics while removing the noise from dynamic PET images. By applying this technique to raw projection data, the noise is removed prior to reconstruction, thus permitting preservation and enhancement of high-frequency spatial information. The method of projection onto convex sets (POCS) has been used to form a deblurred sinogram by utilizing additional detector wobble motion measurements and prior knowledge of the PET system's spatial response function. However, in low-count dynamic studies, this results in noise amplification. The combined methods of PCA and POCS have been applied to dynamic monkey neuroreceptor PET studies to reduce the noise and improve the spatial resolution. Consequently, this improves both the accuracy and precision of the estimation of the related kinetic parameters as compared to the known values obtained from postmortem radioassay. The specific binding and ratio of striatum/cerebellum were improved by factors of approximately 1.6 and 2, respectively.