GPU formulated MLEM joint estimation of emission activity and photon attenuation in Positron Emission Tomography

We present image reconstruction results for Maximum Likelihood Expectation Maximization (MLEM) based joint emission activity distribution- and photon attenuation map estimation from Positron Emission Tomography (PET) data. To our knowledge this is the first MLEM method for such joint estimation in PET. While the proposed MLEM method applies to a family of statistical data acquisition models, we propose a specific model which is appropriate for efficient Graphics Processing Unit (GPU) based implementation. The PET data was generated using a simulated two dimensional 64×64 pixel XCAT thorax phantom, and the joint emission activity- and photon attenuation root mean square reconstruction errors were 3.6% and 1.1% respectively. The typical reconstruction errors had negligible global influence on the quality and accuracy of the reconstructed images.