Analytical image reconstruction strategies for AX-PET data

Axial-PET (AX-PET) concept aims at significant reduction in parallax error and simultaneous improvement in sensitivity and spatial resolution by employing a novel geometry based on stacks of axially oriented long crystals and orthogonally placed wavelength shifter strips. Current AX-PET demonstrator has two modules. Better sampling and larger field-of-view (FOV) are achieved by rotating one of the modules up to ±60°. By using a rotating source, projections can be taken from all possible angles. In this study, we investigate analytical reconstruction strategies for the list-mode AX-PET data. This process consists of pre-processing steps such as construction of 3D sinograms (histogramming) and dedicated geometrical corrections. In histogramming, we used 1, 0.8750, and 0.4375 mm radial and axial sampling with the angular sampling of 1°. The sinograms were constructed by using span9 axial compression scheme. The constructed 3D sinogram data were corrected in order to have homogeneously sampled sinograms and also for the inter-crystal gaps by using transradial bicubic interpolation method. For image reconstruction, we used 3D reprojection (3DRP) with Colsher´s filter (as implemented in Software for Tomographic Image Reconstruction, STIR). The physical phantoms were scanned with the two-module AX-PET demonstrator. The quantitative evaluations were performed by calculating the Full Width at Half Maximum (FWHM) values of the fitted Gaussian curves to the manually drawn profiles across the phantom inserts. The preliminary reconstructed phantoms showed that we are able to resolve the NEMA insert down to 2 mm. Moreover, with the reduced parallax error, we are able to obtain more uniform spatial resolution in the transaxial field-of-view.