Advanced system modeling for fast, iterative, fully 3D positron-emission-tomography reconstruction

Iterative algorithms for PET reconstruction, e.g. Maximum-Likelihood-Expectation-Maximization (MLEM), are well established and often yield superior image quality compared to conventional analytic reconstruction methods. Especially, the incorporation of detector blurring within the modeling of the scanner response can significantly improve the image quality. However, the blurring modeling is accompanied by a reduction of the rate of convergence. Therefore, the statistical accuracy of the acquired data limits the improvements depending on the specific system sampling. We have developed a fully 3D MLEM-reconstruction software for a Siemens/CTI HR+ PET scanner including an appropriate modeling. Here, the true parameters of the detector resolution appear not adequate for the modeling, because strong artefacts in the images are observed. Therefore, we have adapted the parameters of the detector resolution model so that the artefacts disappeared, but that the image resolution was still better than without using the detector resolution model.