CellProfiler - apply improved classification to already analysed data?


I am using rules exported from CPA to classify my cell cycle stages (after segmentation of nuclei and spots & tracking ) and I export csvs with all the measurements including classes (as well as different outlines and labels for the nuclei segmentation).
It is working well but I would like to add an additional class (dead cells) and improve classification now that many more movies have been segmented.
My question: can I apply the new classification to already analysed data or do I have to run the entire pipeline again (which would be super time consuming but well…)

If I import the image labels in the pipeline, the re-classification should be doable while keeping the same TrackObjects_Label, right?

Also, is there a bioligst-friendly way that I could retrain the ClassifyPixel UNET based on my images to avoid oversegmentation of mitotic cells (in particular pro-metaphase cells)? I tried playing with different images sizes but no luck so far…


The best way to retrain them would be to re-run the old data sets through CPA; you can then get a per-object classification table you could use to then link back to your original data.

Sadly, there isn’t a biologist-friendly way to retrain that, no.

ok will try that then - I have the help of the all powerful python gods near me (Phan-Min Son @Phan96844506 and @Anatole Chessel) to help me!