A practical, semi-experimental system matrix for 2-D PET image reconstruction: Comparison with a geometrical model

An accurate system matrix is an important component in statistical PET image reconstruction. Indeed, inaccurate projection-backprojection may wrongly distribute acquired data and thus introduces noise into reconstructed images. Here, we propose a practical procedure for assessing the system matrix, in order to improve image accuracy in any clinical department. To do so, point source acquisitions are performed with a clinical radiotracer (i.e. ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG)). Unlike geometrical models (which have to be pre-computed), the matrix elements in our method are computed on-the-fly during reconstruction. Computation time and storage space are drastically reduced in our approach (OSEMDR), when compared with a geometrical model based on line integral (OSEMGEOM). Acquired data of the NEMA IEC PET body phantom were used to compare images obtained from both reconstruction methods in terms of contrast, noise and the contrast-to-noise ratio. Our results show that OSEMDR significantly increases the contrast-to-noise ratio (mainly by decreasing noise). Hence, our approach (incorporating the PET device’s system response into the system matrix) could be of great potential interest to a clinical department.