LabelOverlapMeasures added

[muschellij2]

I think adding https://github.com/ANTsX/ANTs/blob/master/Examples/LabelOverlapMeasures.cxx would be a great addition to the ANTsRCore package. I need Hausdorff distance for some of my research and it seems like ANTsX/ANTs#611 has indicated that this should be fixed, but ANTsX/ANTs@295ae80 had the hausdorff distance, but it seemed it was immediately removed from @ntustison in ANTsX/ANTs@119dd3b#diff-267dd586075f89a331c8b476d18fe752

Describe the solution you'd like
LabelOverlapMeasures.cpp for overlap measures (including Hausdorff and ASD)

Describe alternatives you've considered
You can do this in R, but seems like the ITK functionality is more optimized for this function.

[ntustison]

I removed the code because of a pre-existing class for calculating the Hausdorff distance. It would be easy to include that class, if needed.

[muschellij2]

I see. I'm looking but, it is possible to calculate Hausdorff or ASD in ANTsRCore? I'm not against trying to adapt this code, but I'm not as familiar with the ANTs codebase to figure out what the easiest path to doing this. Any guidance would be appreciated Best, John

…

On Thu, Nov 7, 2019 at 11:37 AM Nick Tustison ***@***.***> wrote: I removed the code because of a pre-existing class <https://itk.org/Doxygen/html/classitk_1_1HausdorffDistanceImageFilter.html> for calculating the Hausdorff distance. It would be easy to include that class, if needed. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#104?email_source=notifications&email_token=AAIGPLTBXAWLSP5L44Y3KILQSQ75JA5CNFSM4JKJTCD2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDNAS2Y#issuecomment-551160171>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAIGPLTMKPWLAB5O23RK6FTQSQ75JANCNFSM4JKJTCDQ> .

[muschellij2]

Regardless, can close but would be interested in the discussion.

[stnava]

in brief:

max( iMath( segmentation, "D" ) * segmentation )

something like this.

[stnava]

hausdorffDistance <- function( binarySeg1, binarySeg2 ) {
  d1 = iMath( binarySeg1, "MaurerDistance" ) * binarySeg2
  d2 = iMath( binarySeg2, "MaurerDistance ) * binarySeg1
  return( max( c( max( abs( d1 ) ), max( abs( d2 ) ) ) ) )
}
seg1 = thresholdImage( ri( 1 ), "Otsu", 3 ) %>% thresholdImage( 2, 2 )
seg2 = thresholdImage( ri( 2 ), "Otsu", 3 ) %>% thresholdImage( 2, 2 )
seg5 = thresholdImage( ri( 5 ), "Otsu", 3 ) %>% thresholdImage( 2, 2 )
hausdorffDistance( seg1, seg2 )
hausdorffDistance( seg1, seg5 )

probably close to correct

[stnava]

note update - need MaurerDistance because it gives signed distance.