Add ANTs LabelGeometryMeasures for volume stats per label

[binarybottle]

As suggested by @satra and @ntustison in issue @150, we could add volume stats per label computed by LabelGeometryMeasures in ANTs.

@ntustison -- Why would I be getting the following error from the T1 and labeled version of the same?:

jovyan@76e723833dac:~/work$ LabelGeometryMeasures 3 /home/jovyan/work/CBIC/5545866_T1w_VNavNorm/mindboggle123_output/mindboggled/5545866_VNavNorm/labels/freesurfer_wmparc_labels_in_hybrid_graywhite.nii.gz /home/jovyan/work/Desktop/CBIC/5545866_T1w_VNavNorm/t1.nii.gz
terminate called after throwing an instance of 'itk::ExceptionObject'
what(): /opt/ANTs/build/ITKv4-install/include/ITK-4.11/itkImageToImageFilter.hxx:250:
itk::ERROR: LabelGeometryImageFilter(0x3791490): Inputs do not occupy the same physical space!
InputImage Origin: [9.8958282e+01, 1.1097722e+02, -9.7684586e+01], InputImage_1 Origin: [9.8958290e+01, 1.1097722e+02, -9.7684586e+01]
Tolerance: 1.0000061e-06

Aborted

[ntustison]

The first element of the origins are different hence "They don't occupy the same physical space." This is an ITK requirement and somewhere during your processing the origins ended up changing due to, I'm guessing, floating point precision. We end up using CopyImageHeaderInformation during antsCorticalThickness.sh quite a bit to avoid this error popping up.

[binarybottle]

Thanks, @ntustison! It's been a while since I've used CopyImageHeaderInformation. What am I doing wrong in the following?:

jovyan@76e723833dac:~/work$ CopyImageHeaderInformation cbic_t1.nii.gz cbic_labels.nii.gz cbic_labels_t1hdr.nii.gz
jovyan@76e723833dac:~/work$ LabelGeometryMeasures 3 cbic_labels_t1hdr.nii.gz cbic_t1.nii.gz cbic_labelgeometrymeasures.csv
terminate called after throwing an instance of 'itk::ExceptionObject'
  what():  /opt/ANTs/build/ITKv4-install/include/ITK-4.11/itkImageToImageFilter.hxx:250:
itk::ERROR: LabelGeometryImageFilter(0x28f1610): Inputs do not occupy the same physical space! 
InputImage Origin: [9.8958282e+01, 1.1097722e+02, -9.7684586e+01], InputImage_1 Origin: [9.8958290e+01, 1.1097722e+02, -9.7684586e+01]
	Tolerance: 1.0000061e-06

Aborted

[ntustison]

Try

$ CopyImageHeaderInformation cbic_t1.nii.gz cbic_labels.nii.gz cbic_labels_t1hdr.nii.gz 1 1 1

[binarybottle]

Thank you. The output volume looks fine, but why are the label numbers in the csv file all over the place?:

Label

2
24
245
2457
24578
2457810
245781011
2.4578E+10
2.4578E+12
2.4578E+14
2.4578E+16
2.4578E+18
2.4578E+20
...

[ntustison]

I have no idea. You should check the content of your images.

[binarybottle]

Very odd. I get weird labels in the csv file using the command:

CopyImageHeaderInformation ru_t1.nii.gz ru_labels_t1hdr.nii.gz 1 1 1
LabelGeometryMeasures 3 ru_labels_t1hdr.nii.gz ru_t1.nii.gz ru_labelgeometrymeasures.csv

But using the same CopyImageHeaderInformation output file, I get proper labels in stdout using the command:

LabelGeometryMeasures 3 ru_labels_t1hdr.nii.gz`

Hey, does LabelGeometryMeasures compute surface area in the same way as LabelThickness2? Can I generate LabelThickness2 values by dividing LabelGeometryMeasures Volume(voxels) by SurfArea(mm^2)?

[ntustison]

Hey, does LabelGeometryMeasures compute surface area in the same way as LabelThickness2? Can I generate LabelThickness2 values by dividing LabelGeometryMeasures Volume(voxels) by SurfArea(mm^2)?

Yes. I don't ever use LabelThickness2 but rather use LabelGeometryMeasures to keep all that info.

[binarybottle]

Good to know! Thank you, Nick! In that case I have GOT to include LabelGeometryMeasures output with output from the mindboggle123 pipeline, once I figure out how to get sensible label values!

[ntustison]

I noticed you didn't link this plot. Did you ever see it? One of the metrics for our 2014 cortical thickness paper was age prediction. In the paper we only show mean squared error of the FreeSurfer vs. ANTs prediction results for 1000 training/testing data permutations. The plot linked to above was also made including the results of ATITH relative to the two others. I guess you didn't as that would've been something compelling to include in your paper.

[binarybottle]

I definitely don't remember seeing this plot! Thank you for sharing.

I assume that these predictions were made on region-based thickness, not locally (vertex-wise or voxel-wise), correct? If so, how did you generate a single thickness measure per region for ANTs, given that antsCorticalThickness.sh generates an image volume with a different value per voxel?

[ntustison]

I assume that these predictions were made on region-based thickness, not locally (vertex-wise or voxel-wise), correct? If so, how did you generate a single thickness measure per region for ANTs, given that antsCorticalThickness.sh generates an image volume with a different value per voxel?

joint label fusion + DKT labels

[binarybottle]

So that's how you established the labeled regions, but how did you compute a single thickness measure per region from Direct/KellySlater/KellyKapowski? Did you simply take the average across all voxels in the region?