Analyze paired images (GFP & DIC) of yeast cells



I am new to cellprofiler and to this forum. Perhaps this has been discussed before - it would be really helpful to be pointed in the right direction.
I have a paired sets of of images of budding yeast cells. One is DIC and the other is in the GFP channel. Depending on the strain, the GFP channel will have different intensity. Some are so low that they cannot be identified unless you see the paired DIC image. So I was thinking of a following flowchart to quantify these differences.

1.Pair the images.
2. Outline the cell boundaries using the DIC image (get a mask?).
3. Use the boundaries as defined above to identify the regions of interest in the paired GFP image (propagate the mask identified above?)
4. And finally measure the intensity of these regions in the GFP channel.

Happy to provide example images if needed.

Many thanks


Hi @akc,

In theory, the pipeline you propose is reasonable.

The problem you are likely to encounter though is in segmenting your yeast cells using just a DIC image - I think that is probably going to be challenging. You could simplify the analysis quite substantially by using, for example, DAPI staining of the nuclei to identify the approximate centre of each cell and a second stain, either membrane-bound or relatively cytosolic, to identify the extent of each cell.



Hi @djpbarry
Thanks for your response.
You are absolutely correct about the challenges of segmenting yeast cells with DIC images and a nuclear stain would make our program way more streamlined. However, we have a lot of non nuclear stained historical data that we would like to analyze.
We are trying out some cell-profiler pipelines and so far it seems to be working with some error (1-5% overestimation depending on image quality).
But my question turns out to have been more trivial in nature. I think I have figured out how to transport a mask that I build with the DIC images to fluorescent images.


No problem.

To improve results, it might be worth considering segmenting the cells with Ilastik first, then importing the masks into CellProfiler.



HI @djpbarry

I have taken up your suggestion to use ilastik. The part where one trains ilastik using pixel classification went really well. I trained it on about 6 images and it can identify the cells and background easily.
I have about 200 images that I want to process. Iā€™m a bit confused how do I incorporate the results of this ilastik training set into my cellprofiler workflow. Especially, how do I feed this information to help identify primary objects in my pipeline.
The questions, I have are like - should I have to batch process all my images in ilastik and export masks separately for each image which I somehow use in CP. Or will that all happen in CP just using the output of the training set?
Hope my question is clear.
Many thanks for your help.


Hi akc,

the recommended way is to first batch process all images in Ilastik and save the resulting probability maps as tiff.
In a second step, these can be used as an input image in Cellprofiler for segmentation. An example workflow is described in a CellProfiler blog post or in the CelllProfiler Github wiki.

How to incorporate ilastik results into cellprofiler pipeline?