Fusion images of positron emission tomography ( PET and magnetic resonance image MRI are effective to evaluate pathological changes using anatomical information. Registration of PET and MRI based on normalized mutual information (NMI) is widely used , where PET images change a lot depending on not only the choice of tracers but also disease progress ion , but the co registration accuracy has not been confirmed well yet. The aim of this study is to clarify the accuracy of NMI based registration between MRI and PET (amyloid PET and tau PET).
We obtained PET and MRI data from 69 participants (ages from 47 to 87 ) who underwent PET using one of the following four tracers 11 C PiB, 18 F Florbetapir, 11 C PBB3 or 18 F THK5351. T he ground truth (i .e .., a set of registered MRI PET images) was manually made by two experts. For each data set, 40 set of perturbed parameters were given to MR image s . Then each perturbed MR I was registered to the PET image using the statistical parametric mapping (SPM) tool. The differences between the given and the estimated shifts and rotates were calculated as registration errors.
We found the registration errors were around 2 mm in the shift and around 1 deg in the rotation. These registration errors were independent of the amount of the perturbation given.
Although negative subjects of tau tracers showed higher error than other tracers, the registration error was less than PET spatial resolution, which supported a proof of the stability of the SPM.