Registration accuracy of amyloid/tau-PET to brain MRI using modified SPM method

Mutual information-based algorithms, such as those in statistical parametric mapping (SPM), are widely used in multimodal image registration, but they may lose accuracy when tracer uptake patterns diverge from magnetic resonance imaging (MRI) contrast, as in amyloid and tau- positron emission tomography (PET). This study assessed the registration accuracy of MRI and amyloid/tau-PET images using SPM’s default settings and a modified approach (Modified SPM) incorporating a repeated optimization process. Registration accuracy was also examined in PET images without scatter/attenuation correction (uncorrected PET), mimicking PET-only systems lacking structural imaging for attenuation correction. Four PET tracers were analyzed: two amyloid tracers ([11C]PiB, [1⁸F]florbetapir) and two tau tracers ([11C]PBB3, [1⁸F]THK5351). The MR images were manually registered on the PET scans, then misaligned with known perturbations, and re-registered using the default SPM and the Modified SPM. The registration error was defined as the difference between the applied perturbations and re-registration parameters. A voxel size of 2.0 × 2.0 × 3.27 mm3 was used; errors less than one voxel were considered acceptable. The average errors were less than one voxel overall; however, when stratified by PET-negative and PET-positive cases, rotational displacements were observed in tau-PET images with the default SPM only. The Modified SPM reduced these errors and improved the accuracy of uncorrected PET images. Registration accuracy between amyloid/tau-PET and MR images using default SPM was generally acceptable, although larger errors were observed in tau and uncorrected PET images. Incorporating a repeated optimization process (Modified SPM) improved registration robustness, particularly in these challenging cases.