0

I have a brain MRI that contains slices from this angulation:

enter image description here

I was playing with the registration tool for brain MRI called Flirt. The code is available also for Python

I tried to use as reference MRI a brain MRI taken from another patient and with a different angulation. Here a slice taken from this MRI (all the files are in nifti format):

enter image description here

And now, the magic has happened. Giving the first MRI as input to the registration tool and this last MRI as Reference MRI

enter image description here

the result is that the output of the registration tool (Flirt) is a MRI where the slices have this angulation:

enter image description here

So, it seems that the registration tool extracted the angulation of the Reference MRI from the input MRI. This is incredible. How is this possible ?

Paolorossi
  • 57
  • 5
  • Welcome to Stack Overflow! Please take the [tour](https://stackoverflow.com/tour) and read [How to Ask](https://stackoverflow.com/help/how-to-ask). Please [edit](https://stackoverflow.com/posts/62101492/edit) the post to include your own effort into solving this problem. The latter preferably in code, this is called a [minimal reproducible example](https://stackoverflow.com/help/minimal-reproducible-example). – Bilal May 31 '20 at 11:10

1 Answers1

0

What wrapper to you use to convert the dicom image? The dcm2niix wrapper will automatically apply the transformation to orient it correctly. Flirt is a powerful tool that does excellent registration. Many pre-processing pipelines such as fmriprep are using it already. I am interested in how it was able to account for the tilt of the image. Looking at the usage page here https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FLIRT/UserGuide it do not see anything about it accounting for the angle. I wonder if it is just overlaying the activation at each voxel on the MNI template. It would be a good test to orient the nifti image first and then run it through again.
You could create a mask for the unoriented and oriented image and put it through and use a parcellation program to check the voxel count in a specific ROI to see if they are the same for both masks.

I am interested to see what you find!

skebaas
  • 16