Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    bfricker12
    @bfricker12
    Hello, I have been trying to see if I might be able to use cellfinder to automatically count cells in my light sheet images, but I am running into some problems. I think some of them stem from being unclear what the background channel images should consist of. Sorry for the inconvenience, and thank you in advance for any help you could provide.
    Adam Tyson
    @adamltyson
    Hi, the background channel should be a channel without any labelled signal (e.g. just autofluorescence). Typically you might acquire e.g.:
    • GFP labelled cells in the green channel (signal)
    • Autofluorescence in the blue channel (background)
    bfricker12
    @bfricker12
    Thank you, Dr. Tyson, for your help; I really appreciate it. While I don't have an autofluorescence channel immediately, would using a different channel's .xml file result in an error of incorrect syntax for the background input? I copied the .xml file's path directly. If so, I'll just wait until I am able to obtain more images to test out cellfinder. Thank you, and have a good evening.
    Adam Tyson
    @adamltyson
    I’m not sure what you mean, you should only get an xml file for signal channels. What are you trying to do with the xml files?
    Also feel free to get in contact if you need any help. Cellfinder is still a work in progress, and there’s lots to improve.
    Also we could theoretically support single-channel datasets. If you’d like this supported, and could maybe share some test data, you could submit a feature request on the GitHub page, and I’ll see what I can do.
    bfricker12
    @bfricker12
    I assumed the file type for the background image should also be a .xml file so I was imputting that for the -b argument. Is that incorrect? I can contact you via email if that's more appropriate.
    Adam Tyson
    @adamltyson
    Gitter is file. The inputs for both the signal and background channels should be images (or a directory of images). E.g. cellfinder -s signal_image.tif -b background_image.tif
    The xml files store detected cell positions, they don’t have any image data.
    Have you seen the documentation here? https://sainsburywellcomecentre.github.io/cellfinder/main/user_guide/run.html
    bfricker12
    @bfricker12
    Oh, I see! I have been looking at the documentation, but I think somewhere I got confused and thought I needed to change my .czi files to .xml ones for the actual inputs. Would .czi work or should it be .tif?
    Adam Tyson
    @adamltyson
    It has to be .tif at the moment. If you have some .czi files to share, I can look into supporting them too.
    sthitapati
    @sthitapati

    I am trying to do a test run in conda using:
    cellfinder -s C:\Users\Sthita PC\cellfinder_sg\sg.tif -b C:\Users\Sthita PC\cellfinder_bg\bg.tif -o C:\Users\Sthita PC\cellfinder_out -x 2.231 -y 2.231 -z 25

    but repeatedly getting the message
    cellfinder: error: unrecognized arguments: PC\cellfinder_bg\bg.tif PC\cellfinder_out

    Could you kindly point out if I am making any obvious mistake?

    Adam Tyson
    @adamltyson
    There is a space in your username (generally not a great idea), you will need to put each path in quotation marks.
    Also - cellfinder relies on true 3D data (with cells appearing in multiple planes). Unfortunately I imagine you won’t get the best results with 25um z sampling.
    sthitapati
    @sthitapati
    Thanks! I am running trials on already available datasets in the lab. What z sampling would you advise? We'll plan to acquire accordingly in the future.
    Adam Tyson
    @adamltyson
    Is this brainsaw? 2x2x5 works well for us.
    I want to support lower resolution acquisitions though. If you think you need it, feel free to raise an issue on GitHub about it.
    sthitapati
    @sthitapati
    Sure! I 'll get back to you once I run a few trials.
    sthitapati
    @sthitapati
    File "c:\programdata\anaconda3\lib\site-packages\neuro\atlas\atlas_class.py", line 24, in init
    atlas_id = list(data_dict.keys())[0] # assume only one key
    IndexError: list index out of range
    Could you guide me about how to deal with this error?
    Adam Tyson
    @adamltyson
    What is your data? Is it multiple tif planes or just the one?
    sthitapati
    @sthitapati
    Its one tif for signal and one tif for background. Do you advise using separate planes in a folder instead?
    Adam Tyson
    @adamltyson
    Are they multi page tiffs? I don’t think that is supported currently.
    25 replies
    Try a directory of planes.
    sthitapati
    @sthitapati
    Okay! Will do. Thanks
    EmanPaoli
    @EmanPaoli
    Hi Adam I have a question regarding the parameter --threshold, is this a global or local thresholding ?
    Adam Tyson
    @adamltyson
    It's plane by plane
    Chouchi974
    @Chouchi974

    Hi, thank you for your harwork. I may have a problem that i dont know how to solve.
    After the detection of cell and at the beginning of the classification the program stops and it tell me this :
    File "c:\users\guill\miniconda3\envs\cellfinder\lib\site-packages\h5py_hl\files.py", line 173, in make_fid
    fid = h5f.open(name, flags, fapl=fapl)
    File "h5py_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
    File "h5py_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
    File "h5py\h5f.pyx", line 88, in h5py.h5f.open
    OSError: Unable to open file (truncated file: eof = 16910353, sblock->base_addr = 0, stored_eof = 184831728)

    what should I do ? thank you !

    1 reply
    Chouchi974
    @Chouchi974
    cellfinder_2020-06-12_11-25-05.log
    4 replies
    Is it what you asked ?
    rsenne
    @rsenne
    Hi Adam, I'm currently trying to use your program to count cFos+ cells. I'm having an issue using your program where it tells me that no cells are being detected, I'm, just curious if this is related to my input data. The documentation is a little unclear regarding the signal images--should they only include signal or is it ok if they have some autofluorescence? Should i threshold out the background so that only signal appears? Thanks!
    7 replies
    Adam Tyson
    @adamltyson
    Hi @Chouchi974, I'm copying your question here, so that others can find it:
    "HI Adam. I just have a question regarding the generation of the training data. there are a lot of cells that are considerate non-cells by the classification, so I think I have to retrain the network properly. But do I have to generate these training data for the whole brain to generate the Yaml files ? Or I only can do the correction for some plane ? Thanks for your help"
    @Chouchi974 you can generate as much training data as you like, but the more cell candidates you choose, the better the results will be. The original network was trained with 100,000 manually annotated cell positions, but if you're retraining the network, then 1000 may work. You only need to click on the cell candidate and press a key to add it to the training data (and you can do a few at a time), so it doesn't take that long to generate the training data.
    Adam Tyson
    @adamltyson
    Just in case anyone wants to use it, we've released all our training data that was used to generate the supplied pre-trained model. See here for details: https://docs.cellfinder.info/user-guide/training/using-supplied-training-data
    Chouchi974
    @Chouchi974
    thanks for your help ! it works fine for me and my result are way more closer then what I was expecting now. I'll try it a bit more but it should be perfect soon. Just a question, is it normal that the training is that long ? I mean, close to 30min per epoch, I had to reduce number of epoch to 20 and I still have good result of the classification. But still, its really long
    Adam Tyson
    @adamltyson
    Hi @Chouchi974, yes unfortunately it takes a very long time. There is scope to improve it, but it will always take a while. I typically train overnight.
    It also depends on the computer hardware you have, GPU etc.
    By the way, I'd love to see what you're doing with cellfinder if you're able to share anything.
    Oh, and if I set up a workflow for it, would you be willing to share your training data? Just the little 50x50x100um cuboids that are extracted, not the full data.
    Adam Tyson
    @adamltyson
    Hi all. Now that cellfinder has become part of the BrainGlobe software suite, this gitter organisation has been renamed to "BrainGlobe", and all questions regarding cell detection in cellfinder should go to https://gitter.im/BrainGlobe/brainreg.
    I am leaving this channel up for now, but it will be removed eventually. Thanks!