Spaces:

andreped

/

AeroPath

like 11

Running

Hello, @tonycwk !

I guess you are aware that we have already published a fully annotated dataset of 27 CTs here:
https://huggingface.co/datasets/andreped/AeroPath

More information about the dataset, the source code for the web app, and tutorials can be found here:
https://github.com/raidionics/AeroPath

Here is a nice Jupyter Notebook demonstrating how to work with the annotated data in python, to get started :)
https://github.com/raidionics/AeroPath/blob/main/notebooks/aeropath-load-dataset-example.ipynb

But if I understand correctly, you would like to run this model on a large set of CTs to produce pseudo-annotations which you would like to refine to enhance your dataset?

Or did you just want the option to download the produced prediction? I can add a button for this in the UI, if you'd like?

Hi André,

Thanks for your quick reply! Yes. I just want the option to download the produced prediction (in NIfTI). At the moment, the 3D object file is not ideal for my work. If you can do me a favour, it will be great!

Regards,
Tony

@tonycwk OK, sounds good.

Could you open a feature request here about this? Makes it easier to track this feature request, as well as inform when it has been fixed.
https://github.com/raidionics/AeroPath/issues

I am a bit busy today, but if I get the time, I will fix this :]

Hi André,

Can I also confirm that the training and validation data is only from ATM22 for this demo?

Regards,
Tony

@tonycwk OK, sounds good.

Could you open a feature request here about this? Makes it easier to track this feature request, as well as inform when it has been fixed.
https://github.com/raidionics/AeroPath/issues

I am a bit busy today, but if I get the time, I will fix this :]

Many thanks André! I just opened a feature requst and am looking forward to your update!

Can I also confirm that the training and validation data is only from ATM22 for this demo?

Are you talking about which data the model is trained on? Or the example sample that is part of the app?

Regardless, we only trained on the model available in this app using the ATM22 dataset.
However, we used a part of the dataset ATM22 dataset as a hold out for validating the trained model.
We also used the AeroPath paper as a second hold out test set, but AeroPath was NOT used for training the model - only ATM22 dataset.

Can I also confirm that the training and validation data is only from ATM22 for this demo?

Are you talking about which data the model is trained on? Or the example sample that is part of the app?

Regardless, we only trained on the model available in this app using the ATM22 dataset.
However, we used a part of the dataset ATM22 dataset as a hold out for validating the trained model.
We also used the AeroPath paper as a second hold out test set, but AeroPath was NOT used for training the model - only ATM22 dataset.

That's the information I need. Thanks André!

@tonycwk As you might have seen. I merged a PR to address this feature request you wanted now:
https://github.com/raidionics/AeroPath/issues/58#issuecomment-2182517602

Have yet to test it in Hugging Face. Perhaps you could do it for me and report if it works? :]

Hi André,

Thanks for your quick update. I just test it in HF spaces. The downloaded nii is for lung segmentation only. Can you change it to airway segmentation please?

Tony

Thanks for your quick update. I just test it in HF spaces. The downloaded nii is for lung segmentation only. Can you change it to airway segmentation please?

If you want to get the airway result instead, you should manually select the "Airway" Task in the dropdown. When it is done, the airways should be available for download.

Can you give it a try and report if it does not work? :]

It works well now! Thanks again!

Tony