Labeling images using QuPath for semantic segmentation

Part of the Tips and tricks series - Number: 28
Download QuPath from:
qupath.github.io/
Download the groovy script from here:
raw.githubuser...

[@jackli5132]

This is especially useful for cell annotations. While using square/rectangle annotations, I typically encountered high background false classification. Would definitely try Qupath later.

[@DigitalSreeni]

Thanks Jack.

[@bedeamadi9317]

Your contents are awesome, I have gained immensely from them. Thanks :)

[@DigitalSreeni]

Thank you very much :)

[@davidw.9711]

great video, it would be amazing if you could run a ML classifier with Qupath with your own project/an example for a research question

[@adamgogacz607]

Hi Sreeni. Thanks for the videos, they're very helpful. One question, are you planning on doing a video on Mask R-CNN using the masks such as the ones in this video?

[@johnnysmith6876]

Amazing content. As always. Thank you prof!

[@DigitalSreeni]

My pleasure!

[@deepalisharma5909]

Sir, please make video on Vanishing Gradient and Exploding Gradient. I have seen multiple videos on these topics but I'm still waiting for your explanation.

[@soroushkargar7645]

Very informative video as always thank you! I have a question, I saw the image you fed into Qupath was in jpeg format. But does not using such format instead of tiff would lead to loss of data for training an ML model for scientific purposes? Especially when using high bit-depth microscopy images. I am curious what would be the difference in outcome of detection when training using tiff and jpeg each.

[@electime41]

Please also teach tagging in MATLAB, how to create a binary mask and save it in any format, such as PNG, etc.

[@niloufarrahimizadeh1480]

Your video really helped me. Thank you!

[@JahidHasan-jx1ef]

Nice content. Professor would you like to make some video on the medical image processing or similar to Research paper and its code if possible. Just to get some understanding of research paper and its coding properly

[@karinaevgrafova2749]

Hello ! Thank you very much for tutorial ! Could you please help me , how to export masks for two classes? For example, one mask but with different pixel values (0-background , 1 - first class , 2 - second class or two binary masks but in different folders ( one for class 1 and another one for class 2)

[@sarangchahakar1519]

Hey Sreeni, Amazing videos. I am able to understand python in a better way. Can you please make a video of how to automatically insert image metadata into the image properties if I have such two datasets? It would be very informative if you could please help with this.

[@DigitalSreeni]

The best way to handle meta data is by capturing it as XML and embedding it with the tiff file. I am familiar with OME-TIFF file that works exactly this way. There are many libraries to work with OMETIFF and my favorite ones are aicsimageio and apeer-ometiff-library. There may be other ways to handle metadata but I am only familiar with OMETIFF.

[@sirpiratepete]

Great tool, thanks for the recommendation!

[@bielmonaco]

Great video professor, as always. But I'm having some trouble exporting the masks. I'm annotating the whole image to train the semantic segmentation models. Even though I'm following your instructions, the masks are coming out completely black without any annotation data

[@zeeshanahmed3997]

Hello Sreeni, your videos are very helpful and informative.. ! I have a question regarding annotations, if we have 3 classes (like x,y,z) for annotation in dataset but in some images we don't have all three classes that occurs but only 2, so should we leave that class?, as after manual annotation, I have to feed all three classes into model. It is like in image we have persons, traffic signals, and roads. So if there are no person in an image but in other images there are persons, so how to annotate them? Like if I annotate the with only 2 classes roads and signals, will it make problem?

[@YOU-sm6qd]

Thanks for all your effort, sir. I'm wondering if you can help us with complex wavelet transform since it's not available for python anywhere.

[@DigitalSreeni]

It is available in python: dtcwt.readthedocs.io/en/0.12.0/

[@YOU-sm6qd]

@@DigitalSreeni very helpful, thank you a lot.

[@YOU-sm6qd]

@@DigitalSreeni We hope you can do a video about complex wavelet transform for images. Thanks

[@huanwang6236]

Hi Sir, thank you for your quality content as always! Quick question: Do you have a python snippet to convert segmentation boundary to labeled mask? I know how to do the other way using `skimage.segmentation.find_boundaries` but is there any function I can use for the other direction? Thank you so much!

[@DigitalSreeni]

If you want to add boundaries as another class for training, you can use some techniques (e.g., distance transform) to identify these pixels and create a class. A few others asked me the same question so I will consider recording a short video on this topic, hopefully in the next couple of weeks. I cannot suggest you an existing library for this task as I am not aware of any but I need to code it myself and try.

[@petebankhead]

@@DigitalSreeni You can also export boundaries directly with a different label. The official docs contain lots more scripting recipes for export in many different ways: qupath.readthedocs.io/en/stable/docs/advanced/exporting_annotations.html#individual-annotations (see the bit with 'setBoundaryLabel')

[@sophiez7952]

DEAR SIR, CAN THIS SOFTWARE TAG TIF IMAGES WITH COORDINATES? WHAT IS YOUR OPINION FOR ME?THANKS

[@dhbpranav1]

Hi Sreeni, could you do a video on napari as well?

[@DigitalSreeni]

Sure, will do.

[@lailamagid8862]

thank you so much for ur very elaborated videos , I need help with qupath when I upload my images the pixel size is being unknown in the identification button of the image , so when I apply the positive cell detection it doesn't work unless I resize my image into KB rather than MB which compromise the image quality .I'm a beginner user so help is much needed when I did the pixel training the save button is not active how can I use it how can I download the script

[@DigitalSreeni]

Your questions seems to be related to Qupath so I recommend contacting their support. I am not knowledgeable of Qupath beyond using it a a tool for certain tasks.

[@lailamagid8862]

@@DigitalSreeni thank u so much for ur response ; I’m a pathologist ( lecturer of pathology faculty of medicine Ain Shams University Egypt ) with zero experience in coding but very interested in image analysis . I believe my question is very basic looking and might be trivial for any one wz any basic experience 🙈the idea of image pixel and how to identify it in the program . I did try the discussion forums but didn’t get any answer to help solve my easy looking problem . I’m sure ur time is very precious ,was wondering if u can put me in touch with anyone who can have any practical experience with the program and would be willing to give me some time on a zoom practical questions 🫣thank u so much

[@DigitalSreeni]

I don't think there is any question that is basic. Even a 'basic' question like 'why humans do not have fur' can lead to uncovering interesting answers. Anyway, the best way to get help with opensource scientific image analysis software is by posting on this forum: forum.image.sc/

[@victordiasteixeira1694]

Top content! Thanx

[@DigitalSreeni]

Glad you liked it!

[@noorfadzilah1293]

Thank you very much! One question, is there any prerequisite installation for the groovy, as I tried to run the code, but it was disabled in the menu.?

[@greeshmahari5738]

Same happened for me too.

[@jeromeh.3069]

You have to rename the script ".groovy" and not ".txt"

[@yikes1pd]

Thanks. Another way is just copy and paste it on new script and hit run.

[@CreatingUtopia]

Run is grayed out in groovy script what to do sir?

[@CreatingUtopia]

I used polygon method to annotate does that make difference

[@mshz88al]

is it possible to use QuPath for object detection annotation?

[@DigitalSreeni]

The QuPath exported masks are indeed ready for object detection as each object is given a unique pixel value.