[MINC-users] MINC to DICOM

[Andrew Janke]

On Fri, Oct 3, 2008 at 11:07 PM, Alex Zijdenbos <alex at bic.mni.mcgill.ca> wrote:
>>>  <minc> == dcm2mnc( mnc2dcm( <minc> ) )
>>
>> I doubt it.  Header info will certainly be lost unless we add our own
>> custom dicom tags for history patient name, acquisition, etc.
>
> What does MINC add to a header that wasn't in original dicom?

History, provenance (MINC unique ID), unified slice scaling
(sometimes), real to voxel intensity mapping, real valid ranges, slice
intensity ranges. That's what I could come up with just now without
vast amounts of whacking neuron together, I am sure there are more.

> so why would
> it be difficult to put things like patient name back where they came
> from, in pretty standard dicom fields?

Jon used bigger words than I did on this one so I will defer to him.

> Are you talking about producing an output file in format X and that
> output being valid following the spec of X; or of being able to run a
> forward conversion followed by a backward conversion and getting the
> same result you started with? I was talking about the latter, which
> the minc<-> nifti converters definitely don't do. For the fun of it,
> try
>
> $ mnc2nii /usr/local/bic/share/mni_autoreg/icbm_avg_152_t1_tal_lin.mnc test.nii
> $ nii2mnc test.nii
> $ mincdiff  /usr/local/bic/share/mni_autoreg/icbm_avg_152_t1_tal_lin.mnc
> test.mnc
>
> and see how identical they are :)

If you look down, I reckon your toes are smoking....  You my friend
just got bitten by one of the shortcomings of NifTI (slice by slice
scaling). This is somewhat important if you want to do things like
voxel loop or averaging thousands of files with something less than a
PetaByte of RAM.

What you have done above is pretty much just dumbed down the MINC file
to Nifti-1/Analyze level by removing slice scaling. And therein lies
the folly of expecting this to work all the time as the assumption is
that both file formats can hold the same amount/quality of information
when usually they dont.

> And what is worse, on my box (with a
> fresh install of mincbundle of this morning), register doesn't even
> want to load 'test,mnc'....

Lost me there, works for me I think the range thing is biting you.
-scanrange or a bit of post-processing with mincmath -clamp should do
the trick.  The problem again being that Nifti does not support
something like valid-range that I know of. The old ANALYZE/SPM style
calmax and calmin are not used in Nifti

> (b) we are already providing converters that don't
> live up to the forward/backward conversion expectation.

Correct. They do not live up to incorrect expectations. :)

> Yes, I know the answer is "if it is so easy, why don't you do it".
> Touche' on that one :)

Hey! I just wrote you a mnc2dcm.


a