At first glance, regarding identifying RBCs from sickle cells: You could do a binary fill holes operation and then subtract the previous image from the result - you'd have only the small "holes" evident inside normal RBC's shown in your example. Then remove any labeled regions in contact with those holes. You'll probably have a few spurious ones left over, and this might accidentally remove a sickle cell here or there (can't see any in the example, but I'm sure it's possible). However, with minimal tweaking this would remove most normal RBCs from the thresholded image shown. Speaking in the abstract sans example code, here, but it seems like you could get pretty far this way. I'll let someone else chime in about separating cells in contact. Josh On Tuesday, December 9, 2014 1:58:43 PM UTC-6, Claiborne Morton wrote:

Hello Scikit community,

I'm looking for help using particle descriptors to identify sickle cells in the attached image. As you can see the sickle cells are the long, thin cells. I have two issues, the first is that in some cases the sickle cells are in contact with other healthy cells. I am trying to find a way of separating (or water-shedding) the cells so that each "particle" is actually one cell. The second issue is that I am dealing with is trying to find shape descriptors that will allow me to accurately distinguish the sickle cells from the healthy cells.

I used an eccentricity filter on the original image to remove all of the cells with eccentricity less than 0.6. Making this any higher results in removal of sickle cells. What other descriptors might be used for further differentiation?

Thanks, Clay

Hey Clay, Application is everything. =) Depending on your downstream question, one approach might be preferred over another. If your goal is to count normal vs sickle cells, *and* you want to do so often, you could: - use regionprops to get *a lot* of features from your data. There's eccentricity but that'll pick up those clumps of normal cells. You can also compute Euler number (the number of holes in a shape), area, and maybe solidity.  - then run the regions through clustering in scikit-learn. I think you'll get clusters corresponding to sickle cells, normal cells, clumps of 2 normal cells, clumps of a normal and a sickle cell, etc. - then count based on the number of regions in each cluster times the number of cells corresponding to that cluster. Of course this might be too complicated... But simpler approaches will probably have a significant error rate (as you have noticed). Juan. On Wed, Dec 10, 2014 at 7:01 AM, Claiborne Morton <claiborne.morton@gmail.com> wrote:

Hello Scikit community, I'm looking for help using particle descriptors to identify sickle cells in the attached image. As you can see the sickle cells are the long, thin cells. I have two issues, the first is that in some cases the sickle cells are in contact with other healthy cells. I am trying to find a way of separating (or water-shedding) the cells so that each "particle" is actually one cell. The second issue is that I am dealing with is trying to find shape descriptors that will allow me to accurately distinguish the sickle cells from the healthy cells. I used an eccentricity filter on the original image to remove all of the cells with eccentricity less than 0.6. Making this any higher results in removal of sickle cells. What other descriptors might be used for further differentiation? Thanks, Clay -- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image+unsubscribe@googlegroups.com. For more options, visit https://groups.google.com/d/optout.