Exluding nuclei that is not round

cellprofiler

#1

Hi there,

I have optimised my pipeline for nuclei (long time ago), but still constantly refining the process of identification. I think I’ve come to a point where i’m just wanting to exlude certain cells/nuclei that won’t give accurate results.

For example, sometimes CP outlines a nuclei, but splits it in half. So in fact to the human eye, the nuclei is very round, but because of different intensity levels, CP splits it in half. So you have a semi circle with one completely flat side.

How do i go about exluding this certain outline and those that just seem like a very random shape?
I have tried the measureobjectareashape and filterbyobjectmeasurement

what parameters do I need to set?

thanks in advance


#2

Hi Tim,

The same general procedure I described in your previous question applies here as well. To use FilterByObjectMeasurement, you need (a) a Category of measurement, (b) the image that the measurement was taken against, and © a particular Feature of that measurement Category.

So to pick up round objects based on the MeasureObjectAreaShape, you’ll want to use AreaShape as the Category and one of the appropriate Features generated by that module. Each Feature is represented by a number and it is this number you use in FilterByObjectMeasurement (see the help for MeasureObjectAreaShape for the details on the Features and the corresponding numbers). You may want to use the eccentricity or the form factor to exclude objects based on roundness.

Still, I do wonder whether you can better your results but increasing the amount of smoothing in IdentifyPrimAutomatic or using shape rather than intensity to segment the nuclei/cells. Can you give an idea of what you’ve tried?

Regards,
-Mark


#3

Hi Mark,

I’ve tried the form factor feature and that seem to work great! Thanks. However the eccentricity did not work. How should I go about inputting the min/max numbers for that in the filterbyobjectmeasurement?

I have increased my smoothing of nuclei, and put it to shape to distinguish clumps. This seemed to have worked far more better than what I had before. Although I have used the “test” mode many times, it always seemed to have been intensity/intensity to be best for me.

Thanks heaps


#4

Glad to hear that you’ve made progress!

Eccentricity is a value that ranges from 0 to 1, where 0 is a circle, and for [0,1], the shape is an ellipse of some type. The Help for the module describes it in more detail. However, since the measure is non-linear, the appropriate values may be non-intuitive for filtering.

One note about Test mode (which you’ve already found): It shows the results of the different de-clumping/dividing operations using the values of the other parameters that are currently set, especially the smoothing and maxima parameters. If you change those, it helps to run Test mode again to see if the Test results change as well.

Regards,
-Mark