Splitting cells changes the class

josephinekalm

Hi!

Im working with smart segmentation to count DAB-stained and blue-stained nuclei in tiled pictures. I have two classes and have managed to separate blue and brown nuclei pretty good. My problem arises when I want to split clusters of cells with boundary shape split. The splits work fine, but for some newly split cells, the class changes. Usually it changes from being classed as brown nuclei to blue after the split, even though they still are brown according to smart segmentation (and also look very brown to the naked eye). When I try to split the cells with watershed split, all cells are classed as blue after the split and I have no cells classed as brown in my data table. 

What am I doing wrong?

Best regards Josephine

PierreD

Hi Josephine,

It looks like this might be a bug so I'll file it for further investigation. However I also found that checking the option "Split Objects with Count" allows to keep the classes properly assigned.

Best regards,

Pierre

YuriG

Hi Josephine,

When you split objects the new object get the Active class - the class that is selected in the Classes group of Measure ribbon. You just should select the class you want to have after the split and then split objects.

Regards,

Yuri

JeffK

Hi Josephine,

There is an alternative strategy you can employ within Image-Pro Premier to correctly segment your cells and then correctly classify them as  a secondary step.  In many cases this is a more robust way to analyze these types of images/experiments.

The first step is to use Smart Segmentation and find all of your cells, regardless of which group they belong to.  The idea behind this is that in some cases it is easier to find/split all of your cells first due to the mix of dyes/colors.  Once you have done this you can move on to the second step.

When you count your cells, please make sure in the Measurement Types you have selected all of the color channels, intensity and texture measurement parameters.

At this point, you can use the Learning Classification feature and do a multi-parameter classification and train Premier to correct classify your cell population.  The feature is very effective when you have a cell population that contains a mix of dyes within the individual cells.  This method tends to be more robust and less sensitive to changes.  I use this method exclusively when classifying nuclei. This feature was added to Image-Pro Premier v9.1. 

Please take a look at the Webinar for using Image-Pro Premier for Digital Pathology analysis on the Media Cybernetics website.  The Webinar goes through the method of setting up this type of analysis.

Best regards,

Jeff

josephinekalm

Thanks!

I'll try split with count instead and hope that fixes it for now! 

But regarding your answer Yuri, when I split cells, both classes get split with one push of the button, but since they are different classes to start with, I want them to stay in their respective class. I don't want all newly split cells to be the same class. Or can I split each class separately in some way? 

Thank you guys for answering so incredibly fast!

Best Regards Josephine Kalm