Positron Emission Tomography (PET) is one of the most popular methods for accurate localization and quantification of in vivo concentration of radioactivity. However, in most circular PET scanners, the inherent detector geometry results in a distance dependent and variant point spread function (PSF) in the image plane. Off-centered point sources are imaged with impaired spatial resolution in the radial direction resultng in elliptically shaped points. While some attempts have been made to correct for this unwanted image deterioration through incorporation of the variant detector blurring in the iterative reconstruction procedures, in most cases this effect has been disregarded and the images are uncorrected.
In this paper we present and evaluate, by means of computer simulation, a novel approach for correction of the distance dependent and spatially variant point spread function in a two-step procedure. The projections are first homogenized through a non-linear convolution schema prior to reconstruction. In the second step, in order to compensate for the convolution performed in the first step, the spatially shift invariant images thus obtained are deconvolved using a conventional deconvolution method. The preliminary results show that the homogenization can completely remove the shift variant non-isotropic blurring in the image plane. Resolution of the images deconvolved with Lucy-Richardson deconvolution method was improved from 9 mm to 3 mm with maintained shift-invariance.