Hi there!

I can transformed my DICOM images to NIFTIs (Phillips generated) using dcm2niix, and I get 1 single 4D volume. However, when trying to access 3D volumes through SPM, either in GUI or when calling a function, I get this error:

>> spm_file_split
Warning: QFORM0 representation has been rounded.
> In encode_qform0 (line 27)
In nifti/subsasgn>fun (line 109)
In nifti/subsasgn (line 20)
In spm_file_split (line 36)

Any ideas where the issue might be?
Also, can I maybe avoide the 4D creation when transforming using dcm2nii?

Thanks in advance!

Is this a 4D CT dataset, such as perfusion? The QForm can only store a 9 degree of freedom transform (rotation, translation, zoom each in 3 dimensions). The SForm can store 12 degrees of freedom (it can also include shears). Therefore, it is possible to have legitimate SForms that one must round to store as a QForm. You could think of the SForm as a superset of the QForm. This is why most tools give precedence to the SForm. However, you should be aware that VTK/ITK based tools (Slicer3D, ITKsnap, ANTs) give precedence to the QForm, as they only represent rectangular volumes. If these are MRI scans, I assume the shear is extremely small, and just a rounding error in the DICOMs. However, be aware that CT scans can be acquired with substantial shear. Visual inspection could reveal this. If these are CT scans and they are acquired locally, I would inquire whether future acquisitions can be acquired without gantry tilt.

Thanks for the answer.
These are regular fMRI scans, and for some functional sessions the splitting works..
Originally posted by Chris Rorden:

Given it is MRI scans, I strongly suspect this is a rounding error in the originating Philips DICOM dataset. It is unfortunate, but the use of ASCII text with only a few decimal places make rounding errors common when images are acquired with a rotation relative to the scanner bore. In this case, dcm2niix faithfully reports that values reported by the DICOM header, and there is no perfect way to store this as a QForm. If I am correct, this reflects a limitation of the DICOM format, and Philips use of very low precision for the spatial coordinates. Since the warning does not reveal the magnitude of the error, you may want to visually inspect the data to ensure all is well. You may also want to contact your Philips research collaboration manager and lobby them to store spatial coordinates with more precision. In practice, with the subsequent non-linear normalization I doubt these tiny rounding errors have any influence, as the registration algorithms will find the same minima regardless of tiny difference in starting estimates.

https://www.jiscmail.ac.uk/cgi-bin/webad...

Dear Chris,

thank you this is helpful!
Do you have any ideas on how to spilt the 4D file to single 3D files at this point tho?

Is there any reason why you want to split the data into 3D? The 4D representation makes the order explicit and is required by many tools in our field. SPM has supported 4D files since at least SPM5. You could always write a simple SPM/Matlab script for this, but the easiest solution would be fslsplit.

When trying to select 3D files in SPM, it just won't give me this option (see image). SPM just does not recognize individual volumes. I will arrempt your suggestion!

I am unable to replicate this behavior. Feel free to send send me an email (not a forum post) with a link to a dataset on Box, DropBox or GoogleDrive. It would be best to see the DICOM data that generated this dataset.