Object classification over multiple channels


I would like to create a feature, which consist of dividing cells (high intensity in dapi) and a cytoplasmic marker. So the total feature I would like the classifier to recognise is two high intensity nuclei (just divided) and a stain between in another channel. I would like to pick these out from all the normal nuclei as well as dividing once that don’t have the cytoplasmic staining. But I am getting stuck on how to train this over two channels. Any input would be appreciated.

Hi Fredrik

My apologies if I am not understanding your question. But if you are simply trying classify three cell types (where the third is “normal”) and you are not trying to define the features of the classifier then Ilastik can do this easily. However the only limitation is that Ilastik does not provide a lot of control over the model creating or feature it chooses to weight in the classifier.
If you are trying to create your own classifier, ie not a machine learning classifier, then you can do this from Ilastik object measurments or you can place the binary object into a program that allows for such control. In the second case I would recommend cell profiler as it allows for much more regulation over is measurments.
If you are asking to have more control over a machine learning object classifier, then Ilastik is not the best as it classifier is done in the background. I would then suggest taking Ilastik binary into cell profiler, measure your objects into a database and then using cell profiler analyst to highly regulate the classification.

I hope one of these helps.

1 Like