Comparison of PET reconstruction resolution recovery and post-reconstruction deconvolution for partial volume effect correction (PVC) in PET

Objectives In this study we compare an OSEM algorithm incorporating measured point spread functions (PSF) modeling for resolution recovery, and two post-reconstruction methods for PVC in PET.

Methods The first methodology (Alessio et al, IEEE ISBI, 2008:1315) uses measured PSF, which blur among radial bins and are spatially variant in radial location, incorporated into a 3D-OSEM algorithm. Post-reconstruction PVC was applied using two different regularized 3D iterative deconvolution algorithms (Boussion et al, EJNMMI, 2009, 36:1064 / Kirov et al, PMB, 2008, 53:2577). The comparison was performed on both Ge68 PET/CT acquisitions of the IEC phantom with different configurations (acquisition duration, post filtering, number of iterations) as well as 10 clinical 18F-FDG whole body studies. A qualitative and quantitative comparison of the corrected images with those reconstructed using a standard OSEM reconstruction was performed.

Results The deconvolution algorithms showed superior results in terms of resolution recovery and quantitative accuracy, despite a slightly higher (<10-15%) level of noise introduced in the corrected images. For phantom spheres <2cm the deconvolution approaches led to higher (15%-35%) activity concentration recovery relative to the reconstruction based approach including post filtering. The phantom results were confirmed by the patient images with the deconvolution leading to an increase in activity concentration recovery of the lesions of up to 30% and an associated noise increase of <10% in the lungs and liver relative to the reconstruction based recovery.

Conclusions Results suggest that performing both reconstruction and post-reconstruction methods offers improved quantitative accuracy, although post-processing deconvolution approaches lead to overall higher SNR increases