Comparison of transmission- and emission-based attenuation correction for TOF-PET/MRI

PET quantification requires accurate correction for the attenuation of 511 keV photons in tissue. In PET/CT imaging, the attenuation map is derived from a CT image. Contrary to CT, a direct conversation from MRI intensity values to attenuation coefficients is not applicable. Hence, attenuation correction remains one of the major challenges in the development of PET-MR scanners. Many studies have shown the feasibility of MR-based attenuation correction in clinical practice. However, some drawbacks remain. Alternatively, methods have been suggested to derive the attenuation map from the emission data or by using an external transmission source inside the FOV of the PET scanner. In this work three approaches based on transmission and/or emission data were evaluated and compared with a simulation study using GATE. In the first approach an annulus shaped transmission source was inserted inside the FOV of the PET scanner. An iterative MLTR-MLEM algorithm was used to reconstruct the attenuation and the PET image sequentially. In the second approach only the emission data is used and the attenuation coefficients and the PET image are reconstructed simultaneously with the MLAA algorithm. Finally, an MLAA+ method is proposed in which both the emission and transmission data are used for determining the attenuation map and the PET image simultaneously. Results show that the use of TOF information in the emission-based approach is mandatory and the algorithm can be improved significantly by including additional transmission data coming from an external source, especially in cases where the attenuation medium is not fully supported by the activity distribution. Additionally, the presence of the transmission data allows compensation for the low-frequency cross-talk between the reconstructed attenuation coefficients and the PET image. The absolute error of reconstructed attenuation coefficients in the lungs, soft tissue and spine was respectively more then 40%, 15% and 9% higher when only emission data was used. Contrary, in the MLTR approach, where the transmission-data is extracted from the emission data using TOF information, misclassifications cause inaccuracies in the reconstructed attenuation maps and higher inter-tissue variance in the reconstructed PET image compared to emission based methods. These effects are also reduced when the attenuation map and activity distribution are reconstructed simultaneously.