Additional flags used to deface 7T data

[ofgulban]

Hi again,

I have made somewhat extensive changes to the interface, these are:

• Using entry point instead of script (command line call is just pydeface now instead of pydeface.py) to have a flexible structure for adding additional options.
• Usage can now be accessed with pydeface --help or pydeface -h
• --force flag to overwrite previous results.
• --template and --facemask optional arguments to allow custom template and defacing mask instead of the defaults.
• --cost argument to select other fsl-flirt cost function options.
• --applyto argument, which accepts multiple paths to apply defacing based on one image to other images.

I have needed these options to deface images with several contrasts acquired at 7T. First example is a set of MPRAGE data:

Original images:

Stuff outside the head (in the posterior region) is a dielectric pad.

After using the default pydeface:

pydeface PDw.nii.gz
pydeface T1w.nii.gz
pydeface SI_T1w.nii.gz
pydeface T2*w.nii.gz

It can be seen that the defacing fails with certain contrasts (T1w, SI-T1w, T2*w). Even in PDw image where it mostly succeeds it removes a small chunk from the frontal lobe.

After changing the cost function to normalized mutual information:

pydeface PDw.nii.gz --cost normmi
pydeface T1w.nii.gz --cost normmi
pydeface SI_T1w.nii.gz --cost normmi
pydeface T2*w.nii.gz --cost normmi

It can be seen that the defacing still fails with certain contrasts however succeeds with proton density weighted image.

After using --applyto argument:

pydeface PDw.nii.gz --cost normmi --applyto T1w.nii.gz SI_T1w.nii.gz T2*w.nii.gz

Defacing mask based on the proton density weighted image applied to other contrasts gives the best result.

---

Second example to show the new additional arguments is a set of MP2RAGE data:

Original images:

After using the default pydeface:

pydeface T1.nii.gz
pydeface UNI.nii.gz
pydeface INV1.nii.gz
pydeface INV2.nii.gz

It can be seen that defacing fails in all images,

After changing the cost function to normalized mutual information:

pydeface T1.nii.gz --cost normmi
pydeface UNI.nii.gz --cost normmi
pydeface INV1.nii.gz --cost normmi
pydeface INV2.nii.gz --cost normmi

Defacing was successful in T1 and UNI images.

After using --applyto argument:

pydeface T1.nii.gz --cost normmi --applyto UNI.nii.gz INV1.nii.gz INV2.nii.gz

Defacing mask based on the T1 (or UNI) image applied to other contrasts gives the best result.

---

I have tested the new command line interface with several 3T-7T data, it seems that everything works as it was (including exceptions etc.) so the previous users should see no difference other than making the command line call as pydeface instead of pydeface.py.

If you decide to merge this, I would recommend to first release the current version (pydeface 1.1) and then merge this PR in case if people still want to conveniently find the old version.

Also, let me know if you have any suggestions about the interface.

[chrisgorgo]

Looks great! Thank you for the contribution!

[ofgulban]

Thanks :)