Geometric accuracy of reconstructed PET images

The registration of multimodality brain studies (PET & MRI) requires geometrically identical properties of the acquired images. In particular, if pixel by pixel manipulations are performed, e.g., for partial volume corrections of PET images, geometrical discrepancies in the images of one modality lead to erroneous results. When registrating MR images, measured with a Siemens Magnetom Vision, 1.5 T, and PET data, acquired with Siemens/CTI ECAT EXACT HR+, it was noticed, that the PET images were smaller in their spatial dimensions. The geometrical properties of the ECAT EXACT HR+ as well as of the GE/Scanditronix PC4096+ PET scanners were investigated subsequently with a special phantom, using 1.8 mm inner Ø hole tubes filled with a ¹⁸ F solution at defined transaxial positions surrounded by cold water. The same phantom was used in the Siemens Magnetom Vision MR scanner, but with empty tubes as contrast to the surrounding water. The pixel size of the ECAT EXACT HR+, using the standard backprojector of ECAT software version 7.03 is 2% too small leading to a shrunken image. When using a new reconstruction software, which is under development at CTI, pixel size comes out to be smaller only by 0.7%. The dimensions for the PC4096+ as well as of the MR tomographs correspond to accurate values (1.000±0.004)