Would we be able to get an update on how this is coming along? We don't need it to function exactly like it does in Slicer 2.7, but as of now it doesn't seem to work at all.

Can you attach a sample label map before and after the remove Islands operation from Slicer2 that works the way you want it to work in Slicer 4?

I'm trying to better understand why this is failing for you. Here is the sequence I tried:

1) load some image data
2) go to editor
3) use the paint tool to add a spot in the middle of the volume in label 1
4) use the dilate tool to make the spot big (dilate 20 times or so)
5) switch back to the paint and pick label 0
6) add a hole in the middle of the big region
7) pick remove islands and apply

What should happen here is that the hole of 0 goes away. Can you replicate this? How is it different from what you are trying to accomplish?

We will check in on this and get back to you today.

I uploaded a standard "tensor mask" dataset that we would like to use the remove islands tool on. As you can see, there are several holes in the mask. Slicer 2.7 does an excellent job of removing all of these holes, even though some of them are connected to the "sea" in other slices. If this is not possible in Slicer 4, that is fine, we will find a workaround, but I just wanted to show you that it doesn't have the desired effect.

Hi -

The file you attached is only the scene description and it doesn't have the actual data. If you could save a .mrb version I could look at that.

But first, were you able to replicate the steps I listed above? Does that not capture the use case?

Or are you saying that what you want is for the island removal to be applied on a slice-by-slice basis rather than in 3D?

What would be a good way to transfer the data to you, since the .mrb file is too large for Mantis?

I was able to replicate your steps successfully. I suppose that being able to use remove islands on a slice-by-slice basis may be the most similar to what we are used to, although that could end up causing problems as well. Either way, it may be worth testing.

1) There is a partners secure file transfer service: http://rc.partners.org/sFTP

2) Since you were able to replicate the simple test case, can you describe how your use case is different? Is it because the islands aren't really islands but only appear to be on a per-slice basis?

Okay I will use that.

I suppose that is correct. I think the best example of why we want to be able to remove islands on a per-slice basis, would be if you look at the scan that I will send to you, you will see that the eye-ball is included in the automatically generated mask ("labelmap"). In Slicer 2.7, running "remove islands" will remove the mask covering that area, whether or not it is connected in a different slice. However, in Slicer4, running "save island" or "remove island" does not have the desired effect.

Here's a prototype of a new tool to address this:

https://gist.github.com/pieper/6176830

Please test r22299 (should be in a nightly build as of Aug 20, 2013).

There is now a new Morphological option in RemoveIslands that will clean up interior and exterior structures even if there is some connectivity.

PNL reports this works well for their data.