 |
 |
> I have a question regarding resampling of high resolution PET images. We > have a Seimens EXACT HR+ scanner. The images are reconstructed with voxel > sizes of 1.88 mm (X) x 1.88 (Y) x 2.42 mm (Z). There are 63 planes with > 128 x 128 matrix in each plane yielding a FOV of 24 cm X 24 cm in plane and > 15.2 cm in the z-axis. > > As a result the images include the entire brain with room to spare on all > sides. I have noticed that the top and bottom (brainstem) of the brain > appears smeared after spatial normalization. I am using the PET.img > template with the MNI bounding box. Spatial normalization parameters are 4 > x 5 x 4 basis functions, 8 non-linear iterations, sinc interpolation and > 2.0 x 2.0 x 2.0 mm voxel size. > > I have tried changing the normalization parameters, but there is very > little difference to my eye. I have even changed the normalization fudge > factor in the spm_defaults.m, but the 0.02 default value seems to be best. > > My questions are: > > 1) Given that the images have slightly greater resolution in the x and y > voxel dimensions than the default normalized voxel sizes, would it be > better to resample at 1.5 x 1.5 x 1.5 mm voxel size ? Providing that you have adequate space on your disks, and your computers are OK working with the larger images, then this is fine. However, the smoothing is likely to eliminate the benefits. > 2) Given the high resolution of the images, would it be preferable to > reduce the extent of the sinc interpolation, and if so, how does one change > the parameters for sinc interpolation. Alternatively, would the other > interpolation methods be better ? There is logic for reducing the size of the sinc kernel if you have high resolution images. I haven't really explored how much benefit (if any) there is in using sinc interpolation with PET images. The noise in the images may be sufficiently high that movement artifacts due to interpolation may not make a great deal of difference. If you write the realigned images, and then spatially normalize these, then I would definately suggest using tri-linear interpolation (for the spatial normalization). However, if the realignment parameters (in the mat files) are combined with the spatial normalization parameters (i.e. the original images are resampled only once), then you may be better off continuing with sinc interpolation. > 3) What parameters for normalization are other people using for Seimens > EXACT HR+ images ? I don't know what other people are using. At the FIL, we used the default values in SPM96. Regards, -John %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%