Assessment of coregistration of PET H215O images in the abdomen using digital phantoms

The aim of this study was to investigate the performance of PET-to-PET coregistration of abdominal H₂¹⁵O blood flow images from different sessions (before and after anti-cancer treatment) on the same subject. Due to the application-specific nature of image registration methodology, the use of preexisting techniques for new applications requires validation. A quantitative approach was developed based on the Zubal segmented human anatomical phantom in conjunction with blood flow biodistribution data to simulate the activity distribution of a typical PET flow study. Known rotations and translations were applied to the resulting flow software phantom. Poisson noise and spatial resolution of the PET scanner were included in the sinograms created by forward-projection and the reconstructed images were coregistered against a non-transformed reference image using three existing methods: minimisation of standard deviation in the pixel ratio (SPR), multi-resolution maximisation of mutual information (MMI), and K-means variance minimisation (CVM). Assessment based on the transformation matrices shows comparable performance in the range of the transformations assessed (0-15° rotations, 0-15 pixels translations). Assessment based on correlation, residual image and ROI-based results suggest an advantage of MMI for the specific application considered