OTSU Mask in jlf module

[butellyn]

The OTSU mask in the jlf module is leaving big holes in the ventricles of the subjects in a study I am processing.

[a3sha2]

Does it affect all the subjects?

[butellyn]

I haven't opened all of the images, but the volumes for the ventricles are similar for all of the subjects, so I would assume so.

[adrose]

This issue is interesting, the calls used for the nasa project were identical to those used in the PNC.

However because the nasa data are of lower quality, the otsu thresholding rm'ed some parenchyma from the intensity images that JLF was run on.

This issue can be rm'ed by adding the -x flag to the call to JLF.

Here is the original call: ${antsPath}antsJointLabelFusion.sh -d 3 -q 0 -f 0 -j 2 -k $keepWarps -t $target -o $outputRoot -c 1 $atlasString

Here is the new proposed call: ${antsPath}antsJointLabelFusion.sh -d 3 -q 0 -f 0 -j 2 -k $keepWarps -t $target -x or -o $outputRoot -c 1 $atlasString

The command can also accept a user provided mask, which could be the antsCT BE mask as well. I think this should be added to the JLF module as a user provided option, unless if there are any objections.

[adrose]

This is an issue because of the intensity values found in the lateral ventricles though! OTSU thresholding rm'ed these values because if thought these intensity values were background. Most likely a combination of noisy BG and large ventricles... V frustrating.

[adrose]

So we are all in agreement that this is an ants version issue?

@a3sha2 @butellyn

[butellyn]

That's what I think, given that the output that I have from an older version of ANTs does not have this problem.

On Thu, Jan 3, 2019 at 4:09 PM Adon Rosen notifications@github.com wrote:

So we are all in agreement that this is an ants version issue?

@a3sha2 https://github.com/a3sha2 @butellyn https://github.com/butellyn

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/PennBBL/xcpEngine/issues/139#issuecomment-451277947, or mute the thread https://github.com/notifications/unsubscribe-auth/AnHohPSmwzlEOWmNPxWWvLpIXz2YLkHTks5u_nGFgaJpZM4Zoo-j .

[mattcieslak]

At the moment there is no masking being applied because there is no -x argument. Otsu masking is never run so the intensity values in the ventricles shouldn't be masked. Can you think of anywhere else where masking might happen?

Maybe we could delete this section: https://github.com/PennBBL/xcpEngine/blob/master/modules/jlf/jlf.mod#L152

It's not present in the old version that ran on the PNC

[mattcieslak]

I was wrong - the version of ants in the container uses otsu as default. I'm going to change the masking to or and let's see if this fixes it