Dear Luca,
Thanks for the effort with the screenshots. Just to ensure, you wonder about the data as e.g. in the last screenshot, some (more) parts of the brain falls outside the red lines. I wouldn't consider this to be a major issue.
1) Have you turned off the interpolation for displaying the screenshots? The interpolated figures can be misleading sometimes.
2) The normalisation parameters are derived from MRI volumes which should suffer from certain geometric deformations not present in PET data, plus there should be some misregistration error during between-modality alignment. A normalisation which appears to be perfect for MRI can thus lead to a somewhat imperfect normalisation in PET data. Depending on thresholding (what does the red line represent) this might be "more visible" in normalised data.
In principle, and if you want to achieve really high-quality registrations, one could try to geometrically undistort the raw images and/or rely on some non-linear registration techniques (coreg is rigid-body) / keyword modality fusion.
However 3) the spatial resolution of PET data is likely poorer than that of the MRI data, and accordingly, the spatial quality of the normalised PET data will remain poorer.
Finally 4) in case you used the new pushward routine, have you resliced the data before applying normalisation parameters? With the old pullback procedure, it was not necessary to reslice during coreg, but for pushforward, this seems to be recommended - although I haven't tried how much the difference would be.
Hope this helps
Helmut
|