Dear experts,

I'm converting T1 images from a 3T Siemens TIM Trio MRI scanner (acquired 2014) with dcm2niix in MRIcroGL (Version from February, 12, 2016). I want to further process these images in SPM12/CAT12 for voxel-based morphometry analyzes. 
Thereby, a question regarding the conversion of dicom to niftis came up: When I compare the nifti-images converted with dcm2niix, they look the same on a superficial level as niftis converted with the spm12 dicom import function or the older version of dcm2nii (the image without cropping and without reorientation). However, when I more closely compare the images, I see several differences: 

- the origin is not the same
- the voxel size is -1 1 1 for the images converted with dcm2niix (as compared to 1 1 1 for the two other conversion versions)
- the image dimension is 192x256x256 for the images converted with dcm2niix (as compared to 256x256x192 for the other two conversion versions)
- the rotation matrix is not the same

I'm wondering whether these differences are relevant and whether the images converted with dcm2niix are correctly read by SPM12. As I am working with SPM12/CAT12, should I use the SPM12 dicom import function? Or are the images converted with dicom2niix usable with these differences (I will not use different conversions for the same study, but preferably I would want to convert all images with the dicom2niix function)?

I would appreciate very much some insights on this matter.

Greetings,
Julia

Yes, these features are expected. For 3D acquisitions the images are reoriented to be roughly orthogonal with the NIfTI space (e.g. sagittal and coronal acquisitions are losslessly reoriented to axial slices). This should have no influence on tools like SPM (which read the SForm and QForm rotation matrices) but does provide more consistency for simple viewers that ignore these transforms. So the image data is reoriented and therefore the dimensions and rotation matrices are adjusted to reflect this orthogonal realignment. Despite the fact that the headers look different, if you use SPM's "DISPLAY" button will show they are still in register with each other. This helpful feature is only applied to 3D acquisitions (T1 MPRAGE, T2 SPACE, etc), 2D acquisitions (EPI-based fMRI, DTI, etc) are not reoriented because the current slice timing tools assume that the order of slices stored on disk matches the acquired slices.

Dear Prof. Rorden,

thank you very much for your immediate reply. 
I take form your answer that SPM12 correctly reads the converted images as the images are displayed correctly (despite the fact that the header looks differently compared to the SPM Dicom import).
Is my conclusion correct that SPM12 reads all images correctly independent of the conversion method (dcm2nii, dcm2niix, Dicom import function in SPM12) and that these images can be further processed in SPM12 without worrying about any coordinate transformations?

Kind regards,
Julia Binder

Yes, SPM uses the NIfTI spatial transforms to reorient the images. It should work regardless of whether you use SPM, dcm2nii, MRIconvert, dicm2nii or dinifti to process your files. Other tools may either ignore these transforms. As an advanced comment, the header stores two spatial transforms QForm and SForm). Some SPM processes (e.g. coregistration) will manipulate the SForm. Be aware that SPM/MRIcroGL/MRIcron gives precedence to the SForm, while other tools (e.g. those built with VTK) give precedence to the QForm. So after coregistration with SPM the orientation may appear different in different tools. I think SPM is doing the appropriate thing (retaining original coordinates and providing a affine warp to standard space), but other tools do not respect this convention.