Ilastik tracking- Fiji Ilastik HDF5 plugin vs BigDataviewer

Usage & Issues
, , , ,
#1

Hiya,

I want to track cells and within those cells quantify the number of spots.
Channel 1: DNA for cell segmentation and tracking
Channel2 : spots to be counted/analyseed in CellProfiler

I am using multiple softwares for different purposes / trials

  1. Ilastik Pixel Classifier first to generate segmentation, and then the Ilastik tracking module.
  2. fiji to analyse the tracks with MaMuT/TrackMate and hopefully correct some divisions/daughters mis-assignement
  3. cell profiler to use the corrected probability masks and tracks (from MaMuut) in channel 1 but count spots in channel 2.

  1. Data Preparation - H5 format
    it seems I have and hdf5 conversion problem.
    For TrackMate/MaMut, I need to use the BigDataViewer to export into hdf5 and xml (otherwise I can’t export the tracks in the Ilastik Fiji/Mamut plugin version). So I exported my channels separately using the bigdataviewer
    But I trained the pixel classifier and tracking in IIlastik using h5 data subset from Fiji ilastik plugin

  2. pixel classification - tracking in Ilastik
    I trained the classifier on a subset of my movies that were exported into h5 using the fiji-ilastik export module.
    I then went on to the Ilastik Tracking module which worked nicely.
    I wanted to export in a format compatible for MaMut but it did not work as I did not have the xml file (that needs to be generateed with BigDataViewer - hence point 1).
    So I went back to data I havent analysed yet, converted data to h5 using the bigDataViewer and wanted to batch process them using the Ilastik Pixel classifier, but I got an error message

    image
    image859×358 24.1 KB

Is this because the classifier was trained on an h5 image that came from a different source (Ilastik Fiji plugin vs BigData Viewer)?

  1. export of probabilities map /tracks to Cell Profiler
    After checking the tracks of channel 1 with trackmate (and correct them??), I would like to use CP to analyse the 2nd channel.
    image
    image1062×782 77.3 KB

    Would that be the correct export setup (then used for batch files)?
    Would it be doable to export tracks and object IDs

On a side note: I have just been playing with Ilastik for the past day and am really pleased with how well it all works!!
Cheers

Debbi

#2

Hi @Debora_KellerOlivier,

your pipeline sounds very interesting :slight_smile:

currently, the advice that I have to give there is to start off with the data exported from big-data viewer to make sure the dimensionality is correct, and not mix the ilastik-fiji export plugin with that.

this is a sad bug, that is still present in the current ilastik version: https://github.com/ilastik/ilastik/issues/1854
As a workaround for this you could use the ilastik headless mode where you control ilastik via the command line to apply a pre-trained project to new data. Would be happy to help setting that up.

Concerning the spot counting I would hope that CellProfiler is fit for the task, but I don’t know, let’s hope someone knowledgeable stumbles over this thread.

Cheers
Dominik

#3

ouh headless sounds fun! I’d be happy to try as it seems to put a lot of memory in cache if I load my files directly into the GUI (300Mb-1Gb).
How can I run it in headless mode? I guess some python command line (sorry I am a biologist…)

I’ll try the BDV and see what I can get! I am seeing “The Man” (aka JYves) tomorrow so will surely get some useful advice there too.
Cheers
Debbi
Debbi