search

bkainz / fetalReconstruction

GPU accelerated source code for motion compensation of multi-stack MRI data
59 stars 28 forks source link

Error: "Something were wrong - sigma: max, min = nan" #9

Open Utena opened 7 years ago

Utena commented 7 years ago

Dear Kainz,

Thank you very much for your great work on MRI reconstruction! ;) I have installed your last version of the code successfully and I have obtain a very nice reconstruction of 3 MRI stacks. However, I have tested your code on different MRI images (acquired using the same procedure as the first) and the following error appears on my terminal:

Error: "Something were wrong - sigma, std: max, min = nan"

And the final result is an empty reconstruction... :(

Do you know what happens?

Thank you very much! Kind regards,

Jordina

bkainz commented 7 years ago

Hi, some scanners (or K-space recons?) produce image intensity spikes (often single max short values). If one of your stacks contains such a problematic pixel then the EM algroithms will start to produce NaNs at some point. Can you try to normalise and clamp the value range of your stacks to some reasonable and uniform range, e.g. 0-3000 ? Another possibility is that the initial overlap is very bad or non-existent, for example when the image transformation in the header has been messed up by some conversion tool (e.g. DICOM to Nii). Double check if your stacks overlap correctly with some tool like for example the MITK viewer.

best wishes, Bernhard

Utena commented 7 years ago

Hi Bernhard,

I have normalized my data between 0 - 3000 and I have checked the overlap of the MRI stacks... but the following error still appears:

...... Iteration 0. 0 Something went wrong: sigma= nan mix= nan 1 Something went wrong: sigma= nan mix= nan 2 Something went wrong: sigma= nan mix= nan 3 Main loop end masking volume Masking done Sync done. ready for more... .....

I do not know what is happening, it seems that the input data is correct :(((( I can easily reconstruct some MRI, but with others I get the error above. It is strange..., I applied the SAME pre-processing procedure in all cases. As you mentioned, if the initial overlap is very bad in some of my MRIs, it will be impossible to reconstruct them? After re-checking the overlap of my MRIs using the MITK viewer, it seems that there is room for an initial good registration.

Thank you very much,

Jordina

bkainz commented 7 years ago

can I have a look at this particular dataset? (use email/fileshareing if you can hand them out)

Utena commented 7 years ago

Yes! Please, check your e-mail :)

Best wishes,

Jordina