MR-guided reconstruction of PET data in spinal cord PET/MRI

Accurate PET reconstruction in spinal cord PET/MRI is challenging due to the small size of the structure and interference from background activity. The aim of this study was to establish whether MR-guided PET reconstruction can improve the accuracy of measured uptake in the spinal cord.

The hybrid kernel expectation maximisation (HKEM) algorithm was evaluated on a digital anthropomorphic phantom (XNAT), and an implementation of a modified asymmetric Bowsher's prior incorporating both PET and MR data was evaluated on clinical test cases. The methods were compared against commonly used algorithms OSEM and Q.Clear.

The results demonstrated that the two algorithms lead to an increase in measured [₁₈ F]FDG PET tracer uptake in the spinal cord. Comparison to ground truth indicates that the improvement is insufficient to remove the bias in this small structure.

With care taken to optimise for the desired application, novel PET image reconstruction algorithms using PET and MR data to inform iterative image updates lead to improved quantification and improved image quality compared to OSEM. Further work is needed to investigate the optimal parameters and identify strategies to reduce residual bias.