Видео

I assume you are trying to use StarDist? If yes look at the code part halfway through this video (e.g at 21:30). At some point you have to specify on which channel you want to work with. In this example it's 'DAPI'. It may be different for you, you can look for the channel names in the Brightness/Contrast tool (the half black circle in the toolbar next to the drawing tools). If you are asking about running a classifier you have trained, make sure you apply it on objects detected in the same way as the objects used for the training. For instance the StarDist script doesn't always include the "measureIntensity()" command, meaning a classifier trained on detections which have an Intensity feature will not work on detections that have not this feature. One common issue is that the Analyze>Cell detection command adds the intensity features by default, but not the StarDist code snippet. Another common mistake is trying to apply a classifier trained on images with certain channel names on images which have other channel names (for instance, DAPI became Alexa405nm or something similar). There are many potential issues like that, the key being to try your best to stay consistent.

@@UniversalBuilder Hi, Thank you for your reply. I am having problems training classifiers for different channels. I do not understand why despite making it to annotate cells to identify classifer, command live update is not working for me

All the available parameters are listed here: qupath.readthedocs.io/en/stable/docs/deep/stardist.html You can change the cell expansion radius, but not the nuclei. I suppose you could filter out later nuclei out of your target range. Something like that: forum.image.sc/t/select-delete-objects-that-are-above-a-specific-area/38229

Stardist is quite sensitive to the cell (or round object) size. A good trick is to scale the image according to what stardist expects. Example of the issue here with a blob image scaled up 8x : ibb.co/hLh9Hch . Another way to improve woud be perhaps to try to do a better deconvolution. If not possible, try to resort to pixel classifiers plugins like weka or labkit to create a mask of the regions you want to exclude, then use stardist on the masked copy of the image

@@UniversalBuilder Thank you for the suggestions! 😎

This wasn't my sample so I'm not 100% sure but this was probably a simple hematoxylin and eosin labeling.

If you are referring to the acquisition with the slide scanner it was probably the 10x plan apo

Sorry for that. Since then I invested in a decent mic and the sound should be much better in the next videos. I'm a bit short of time these days so I had no updates for a long time but I'm planning to get back on track soon. Thanks for watching !

Thanks ! You can find some answers here : imagej.net/plugins/stardist And there's a FAQ (that also applies to training models and the python version) here stardist.net/docs/faq.html In general, start with the suggested parameters for the model. If it doesn't work well, try first to check the size of your images to see if they need to be resized to get a proper segmentation. The rest depends on your images, so you might have to check one parameter at a time and see how changing it improves or not the results.

Hi, it's possible but I wouldn't suggest using Fiji for that. I would use something like Imaris, or maybe try other tools like APEER, ... In order to enable 3D segmentation with StarDist and Fiji you have first to install and configure a proper Python environment and the StarDist full package. Then you can use a wrapper to connect from Fiji to your Python installation. There are instructions here : github.com/BIOP/ijl-utilities-wrappers The rest of the workflow should be quite similar. As of making a video for that, I'll see if I have time but no promises there, as I'm really too busy in the moment :(

@@UniversalBuilder Thanks for your suggestion I will try to do

Slightly late to answer (sorry didn't see this one), Yes it's the number of particles analyzed.

Thank you ! I will have a look at this question, but I'm currently waiting the release of the 0.4 version to check if there's something new on that side. If not, I will record a video to show how to do that from scratch

@@UniversalBuilder Thanks a lot!

Hello Marion, just a quick update: QuPath 0.4 is finally out! Regarding Tensorflow, it's as I suspected: everything was rebuilt and the past extension is now obsolete. Building QuPath from scratch isn't needed anymore apparently, or at least it seems much easier. Please look at this thread from the developper : forum.image.sc/t/qupath-v0-4-0-now-available/74887. Good stuff :)

@@UniversalBuilder hello Yannick, thanks for your commentary! I have also seen this new version and it is much easier to use Stardist with Tensorflow and that is a good thing!

I think you could probably do a variation of the tutorial above and add another step to discriminate between the fibers, but it's difficult to say without knowing what you consider as "positive". Is it a change of size or shape ? Of staining color ? Of texture ? Depending on that you could use tools to get information from the ROIs created using this tutorial to create a new heat map based on the measurements, and threshold it to keep only the fibers you estimate as positive. The Biovoxxel and MorpholibJ plugins have tools for that if I recall correctly.

@@UniversalBuilder , thank you very much for your response!