Issues with proper overlay of segmentation

[ryancrossbow]

Dear Dr. Bonaretti,

I have been working with pyKNEEr and I have had some success and some difficulty. Of about 20 images that I wanted to automatically segment, about half of them were done correctly.

For the other half I am having two consistent issues. The issues are that the automatic segmentation: 1.) will not properly overlay on the T2 structural image, and 2.) it seems that the segmentation of femoral cartilage is not always accurate. It looks as if the segmentation is in a different 3-D space than the T2 preprocessed .mha file. For example, I performed an automated segmentation, with pyKNEEr, on BD and it generated a segmentation that is not spatially correct (Image 1). To try and get pyKNEEr to properly segment the image I then created a ground truth segmentation (Image 2) and placed it into the proper new_subject folder.

Therefore, I changed the ground truth segmentations into the proper 'reference.mha' files. :
ground_truth_fc --> reference_fc.nii
ground_truth_f ---> reference_f.nii
preprocessed_image ---> reference.mha

Image 1

Image 2

There were no issues created by me using nifti files in place of mha files. Also, the code was able to run and it generated a more accurate segmentation of the femoral cartilage (fc); however, the segmentation still was out of frame (Image 3).

Image 3

I ask whether or not this has to deal with the fact that I have not changed:
reference_fc_15.mha
reference_f_15.mha
reference_fc_levelSet.mha
reference_f_levelSet.mha

When I open the the ground truth segmentations (reference_f.mha and reference_fc.mha), pyKNEEr generated segmentation, and the reference_fc_15.mha and reference_f_15.mha files in ITK-snap I see that my ground truth segmentations always exist within the reference_x_15.mha space. Furthermore, I notice that the pyKNEEr generated segmentation always exists within the reference_fc_15.mha and reference_f_15.mha.

Do you think this is the issue of why the segmentation is not overlaying correctly? If not, do you have any other suggestions.

Please let me know, thank you!