I was also wondering if there is any way to code everything on a notebook rather than being back and forward with Cell Profiler and Ilastik. I am sure there is and I was wondering if you have already tried to do so and, if yes, if you can share it š
In-house we have a development setup where we can directly orchestrate everything from a jupyter notebook and even deploy it on our cloud infrastructure, essentially by wrapping around the command line interfaces of cellprofiler/ilastik.
However such implementations are quite tailored to the local infrastructure, operating system, how programs are installed etc, so I am a bit hesitant to share it before we do not have it well streamlined and documented.
I also used to have such sections in the tutorial but based on feedback it seemed to confuse more than actually help :/
Also: While it is possible to make a rough, completely automatized segmentation just based on nuclear and total intensity stain alone, whenever you want leverage pixel classification, a manual step jumping to ilastik is not really avoidable except if very standardized pannels/samples are used such that the training can allways be reused.
Question by email:
In-house we have a development setup where we can directly orchestrate everything from a jupyter notebook and even deploy it on our cloud infrastructure, essentially by wrapping around the command line interfaces of cellprofiler/ilastik.
Here some pointers how this can be done: CP: https://github.com/CellProfiler/CellProfiler/wiki/Adapting-CellProfiler-to-a-LIMS-environment#cmd Ilastik: https://www.ilastik.org/documentation/basics/headless.html Wrapping command lines interfaces from jupyter notebooks: http://mmcdan.github.io/posts/interacting-with-the-shell-via-jupyter-notebook/ Gc3pie: https://gc3pie.readthedocs.io/en/master/ (a package to orchestrate batch execution of command line programs on clusters)
However such implementations are quite tailored to the local infrastructure, operating system, how programs are installed etc, so I am a bit hesitant to share it before we do not have it well streamlined and documented. I also used to have such sections in the tutorial but based on feedback it seemed to confuse more than actually help :/
Also: While it is possible to make a rough, completely automatized segmentation just based on nuclear and total intensity stain alone, whenever you want leverage pixel classification, a manual step jumping to ilastik is not really avoidable except if very standardized pannels/samples are used such that the training can allways be reused.