Joint reconstruction of activity and attenuation in non-TOF PET using a synergistic prior to enforce structural similarities

This study investigates the use of a synergistic edge-preserving prior for a maximum-likelihood reconstruction of PET activity and attenuation images. The effect of the prior was tested on different reconstruction methods: MLAA and our algorithm MLAA-EB-S. Simple simulations on cylindrical phantoms with an insert showed that the presence of a synergistic prior enforces cross-talk related edges when the images are reconstructed with non-TOF MLAA and increases the bias in both the reconstructed activity and attenuation images. However, benefits appear when the joint problem is less ill-posed, as in the case of our method MLAA-EB-S, which exploits scatter information to improve conditioning. Overall, MLAA-EB-S with synergistic prior achieved the lowest bias and variance in both the activity and the attenuation images amongst the compared methods. Therefore, the proposed synergistic prior could be a useful way for improving the performance of MLAA-EB-S without the need of relying on anatomical images.