[Neuroimaging] nibabel.trackvis.read error

[Matthew Brett]

Hi,

On Fri, Sep 4, 2015 at 12:27 PM, C.D. Langen <c.langen at erasmusmc.nl> wrote:
> Hi Matthew,
>
> Thank you for your quick reply. Below are links to two datasets, one
> that failed to be read by nib.trackvis, and one that succeeded. Both can
> be viewed in Trackvis:
>
> https://dl.dropboxusercontent.com/u/57089115/fail.trk
> https://dl.dropboxusercontent.com/u/57089115/succeed.trk
>
> Best,
> Carolyn
>
> On 03-09-15 19:31, Matthew Brett wrote:
>> Hi,
>>
>> On Thu, Sep 3, 2015 at 10:22 AM, C.D. Langen <c.langen at erasmusmc.nl> wrote:
>>> Greetings,
>>>
>>> When I try to run the following line of code:
>>>
>>> streams, hdr = nib.trackvis.read(os.path.join(subjDir, 'dti.trk'),
>>> points_space='voxel')
>>>
>>> I get the following error, but only for a small subset of subjects:
>>>
>>>     File
>>> "/cm/shared/apps/python/2.7.6/lib/python2.7/site-packages/nibabel/trackvis.py",
>>> line 223, in read
>>>       streamlines = list(streamlines)
>>>     File
>>> "/cm/shared/apps/python/2.7.6/lib/python2.7/site-packages/nibabel/trackvis.py",
>>> line 202, in track_gen
>>>       buffer = pts_str)
>>> TypeError: buffer is too small for requested array
>>>
>>>
>>> Someone else had a similar error
>>> (http://mail.scipy.org/pipermail/nipy-devel/2012-March/007272.html)
>>> which they resolved by using nibabel from github. I tried this, but got
>>> the same error.
>>>
>>> All subjects' trackvis files were produced in exactly the same way using
>>> Trackvis, so I am not sure why only a few subjects fail while others
>>> succeed. Any ideas?
>>>
>>> Thank you in advance for your help in resolving this issue.
>> Thanks for the report - would you mind put the file online somewhere
>> so we can have a look?

What seems to be happening is that the last track in the file is
truncated.  It says that it is 120 points long (n_pts field), but
there is only data in the file for 77 points.

I tried reading the file with this MATLAB toolbox :
https://github.com/johncolby/along-tract-stats

>> [header, tracks] = trk_read('fail.trk');
>> tracks(end)

ans =

    nPoints: 120
     matrix: [77x3 single]

>> tracks(end-1)

ans =

    nPoints: 91
     matrix: [91x3 single]

Note that the last track has 120 'nPoints' but only 77 points.  The
previous track has 91 'nPoints' and 91 points, which is what I would
expect.  So I think the file is mal-formed and trackvis is being more
generous than nibabel.  I think nibabel should have a mode where it
passes through this kind of thing.   In the meantime, if you want to
read all but the last shortened track, you could do something like
this:

import nibabel as nib

track_gen, hdr = nib.trackvis.read('fail.trk', as_generator=True)

tracks = []
while True:
    try:
        track = next(track_gen)
    except (StopIteration, TypeError):
        break
    tracks.append(track)

Cheers,

Matthew