1

I have registered two images, let's say fixed and moving are Registered. After registration I want to measure overlap ratio etc.

The SimpleITK has overlap measure filters and to use overlap_measures_filter.Execute(fixed, moving) and hausdroff_measures_filter.Execute() we need to segment the image and we need labels in input. But the image is hard to segment using just thresholding or connected component filters.

Now the question is then how can we evaluate registration accuracy using SimpleITK with just fixed image and the registered image.(without segmentation ad labeling the image)

Subhajit
  • 43
  • 9

1 Answers1

0

If I understand your question correctly, you want the impossible: to have Hausdorff distance measure as if the image were segmented, but without segmenting it because the segmentation is hard.

Dženan
  • 3,329
  • 3
  • 31
  • 44
  • Thanks for your reply. So there are no other method to evaluate registration accuracy without segmentation. right? diff image is not robust for multimodal setting. My concern was , is it possible that, registration without segmentation gives better result than registration after segmentation? – Subhajit Mar 21 '17 at 06:19
  • You can have registration using gray levels. Whether that will be better depends on many factors: quality of segmentation, complexity of shape which gets registered etc. But even if it is not better, it might be close enough. P.S. up-vote my answer if it helped. – Dženan Mar 21 '17 at 15:50
  • Thanks a lot! I am extremely sorry, for now I don't have 15 reputations, so I could not able to up-vote. If possible can you please share your thoughts to my other question http://stackoverflow.com/questions/42943272/image-segmentation-and-registration-using-simpleitk Thanks – Subhajit Mar 22 '17 at 05:16