Counting cell colonies with Icy

[Juliano Faccioni]

Hello all,

me & my lab have been struggling for a while to count cells in colonies from our experiments. We can count them manually with plugins like ImageJ's Cell Counter, but this is of course very time-consuming and monotonous, especially when there may be 300+ cells in a single colony and hundreds of images per experiment.

Icy has been our best tool yet when trying to automate this process, but the results are still below our desired expectations. Plugins like HK-Means do a great job at detecting cells in fluorescent images, but it tends to cluster three or more close cells under the same ROI. Unfortunately we cannot adapt our seeding/growth protocol to work around this, since cells stemming from the same colony always tend to grow close/attached to each other. At best, HK-Means detect around 65% of the cells (comparing the number of counted ROIs to the manual counts). 

We are aware that Icy can't perform miracles when it comes to thresholding/detecting cells, especially when the source material is not well-suited for this type of automation (this is partly why we've given up on automating cell counting of light microscopy - the contrast is just not good enough). Still, we were wondering if anyone has any experience in automation of cell counting, in particular cells clumped together or attached to each other? Maybe using some different plugin or a combination of some plugins/protocols?

I've attached some pictures of a recent t-stack of ours as an example of what we need to count (along with the detected ROIs on the .xml file). Many thanks in advance!

Cheers,

Juliano

[Stephane]

Hi,

Given your image it's indeed very complicated to correctly count the number of cell. I made some tests using HK-Means and couldn't get really better from what you got.
I tested with spot detector, it's not good for segmentation but it's ok if you just want to count cells, i believe you can obtain better results with this one on your image.
On a single frame, i obtain ~220 with HK-Means with best setting while i obtain ~320 cells with Spot Detector (you can see n attached screen-shot).
It's not perfect, lot of merges and sometime separations where you shouldn't have, but i believe it's still provide an acceptable counting result. Note than i filtered any detection below 100 pixels as they were spots we don't want.
You can also try the "Split ROI" plugin but from my tests it doesn't really help here (incorrect split almost time).

Hope that help.

- Stephane