MIL example

[myron]

This adds an example of MIL classification from WSI images.

Let's discuss, which components we can move to MONAI core.
I think we can move MILmodel and ImageExtractPatches transform.
There are more components we can move too (or just keep them here as part of the tutorial)

[bhashemian]

It's better to move this file to pathology/multi_instance_learning/ folder. Thanks

[myron]

I've moved it to pathology/multiple_instance_learning folder and updated the code to import existing components from monai (newly added and fixed components/transforms)

[myron]

There are some utility functions left here, I intend to keep them in the tutorial without adding to monai core:

• AverageMeter
• distributed_all_gather
• LabelEncodeIntegerGraded
• SimpleTileFlip

@drbeh sounds good?

[bhashemian]

There are some utility functions left here, I intend to keep them in the tutorial without adding to monai core:

• AverageMeter
• distributed_all_gather
• LabelEncodeIntegerGraded
• SimpleTileFlip

@drbeh sounds good?

Thanks @myron!
At the end, we have to move all this components to MONA but it is about what the most we can add into MONAI for this release, so let's see what else we can implement till Nov 18.
LabelEncodeIntegerGraded seems to be an important piece, isn't it?

[myron]

LabelEncodeIntegerGraded - a way to embed class (when a class is a relative tumor grade). I've not run into it much previously, so it's probably not very common. We could consider it adding it

[myron]

I've updated the tutorial

• using collate function to reassemble data (from list of dicts)
• adjustments for the changes in transforms

[bhashemian]

@myron could you please remove datalist_panda_0.json, upload it here, and add download instruction in the notebook:

dataset_url = "https://drive.google.com/uc?id=........."
dataset_path = "datalist_panda_0.json"
gdown.download(dataset_url, dataset_path, quiet=False)

[myron]

I've made the changes

[bhashemian]

@myron can you address the unresolved comments so that we can merge this example?

[bhashemian]

Just waiting for #431 (comment) to be addressed.

[bhashemian]

@Nic-Ma, the PR looks good to me. Have your comments been addressed? Do you have any other comment? Thanks

[Nic-Ma]

Paste the comments from Daguang:

1. Do we need “pip install "git+https://github.com/Project-MONAI/MONAI#egg=monai[tifffile,imagecodecs]"”? This is a MONAI tutorial and people should have MONAI installed already.
2. Can you add some figures for the curves? Training loss and validation curve.

Thanks.

[bhashemian]

@myron

For the dependencies, you can follow the “Install dependencies" section of other pathology notebook: https://github.com/Project-MONAI/tutorials/blob/master/pathology/tumor_detection/ignite/profiling_camelyon_pipeline.ipynb

First, you can check if MONAI is install or not, and then install it only if the user doesn’t have MONAI installed. Also you can do the same for the individual dependencies, so if the user has MONAI installed but not the dependencies, you can install them only if they are not already installed:

!python -c "import monai" || pip install -q git+https://github.com/Project-MONAI/MONAI#egg=MONAI[tifffile,imagecodecs]
!python -c "import tifffile " || pip install -q tifffile
!python -c "import imagecodecs " || pip install -q imagecodecs

[myron]

I've added the curves

[myron]

@myron

For the dependencies, you can follow the “Install dependencies" section of other pathology notebook: https://github.com/Project-MONAI/tutorials/blob/master/pathology/tumor_detection/ignite/profiling_camelyon_pipeline.ipynb

First, you can check if MONAI is install or not, and then install it only if the user doesn’t have MONAI installed. Also you can do the same for the individual dependencies, so if the user has MONAI installed but not the dependencies, you can install them only if they are not already installed:

!python -c "import monai" || pip install -q git+https://github.com/Project-MONAI/MONAI#egg=MONAI[tifffile,imagecodecs]
!python -c "import tifffile " || pip install -q tifffile
!python -c "import imagecodecs " || pip install -q imagecodecs

I mean, I know how to do it. But what do you suggest to put in README for users? do you want to add all these instructions (seems pretty verbose and complicated). I think we can keep it as is. Thanks
Especially since other tutorials (most), don't specify anything e.g. https://github.com/myron/tutorials/tree/master/pathology/tumor_detection

[bhashemian]

@myron there are few formatting issues that I mentioned. Apart from that it looks good to me.
@Nic-Ma do you have any other comment?

[myron]

done

[Nic-Ma]

Hi @drbeh ,

I see you marked request change which blocks merging this PR, do you have any more comments?

Thanks.