Call Vista3D NIM on NVIDIA AI enterprise / GPU cloud

[aylward]

Fixes # .

Description

Illustrates functions for uploading images to file.io and using their URLs to launch Vista3D NIM on NVIDIA AI enterprise / GPU cloud servers.

Enables researchers with limited GPU resources to take advantage of NVIDIA resources while evaluating MONAI Vista3D.

Checks

• Avoid including large-size files in the PR.
• Clean up long text outputs from code cells in the notebook.
• For security purposes, please check the contents and remove any sensitive info such as user names and private key.
• Ensure (1) hyperlinks and markdown anchors are working (2) use relative paths for tutorial repo files (3) put figure and graphs in the ./figure folder
• Notebook runs automatically ./runner.sh -t <path to .ipynb file>
  • Notebook requires a private API key. Anyone can register to get one, but notebook will not automatically run without one.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[KumoLiu]

Do you think adding two screenshots here would improve the overall clarity and visual appeal?

Reply via ReviewNB

[aylward]

Added.

[KumoLiu]

You may need using ![Figure](Figure.png), the screenshots not shown as expected here: https://github.com/Project-MONAI/tutorials/blob/95eff84dfbadcfba066c3b09e88176d905e1aff7/nvidia_nims/vista_3d_remote_nim.ipynb

[KumoLiu]

Thanks for the PR! Looks good to me.
Put two minor comments.

[ericspod]

Line #5.        invoke_url = "https://health.api.nvidia.com/v1/medicalimaging/nvidia/vista-3d"

I would turn this into an argument for the function, and where you call this function pass in this URL explaining what it is, and maybe suggest other models that would work here too to do something different. I think that helps illustrate what the end point is in the notebook by explicitly using it in a cell rather than being hard coded in a function. At the top of the notebook I would also provide very brief descriptions of what VISTA 3D and NIM are, but very brief next to the URLs you provide for each.

Reply via ReviewNB

[aylward]

Quite a bit of this function is specific to the VISTA 3D NIM and this web interface/URL (e.g., the validation logic and the URL request parameters/arguments). So, I don't think the URL should be a parameter to this function.

However, I agree that a brief description of VISTA 3D and NIMs are needed, and I will add them.

[ericspod]

Line #9.        isotropy = (

Here I'd recommend saving the spacing to a variable: input_spacing = nput_image.GetSpacing() and then use that variable everywhere you're calling input_image.GetSpacing(). This reduces the code a bit and helps with readability.

Reply via ReviewNB

[ericspod]

Line #55.                    return itk.imread(filepath, pixel_type=itk.SS)

if imread can read from a data stream you can use z.open to open streams to files in the zip file without extracting it, for large zips this can be faster.

Reply via ReviewNB

[aylward]

Streaming from zips is already done by ITK (i.e., imread) depending on the file format (e.g., you can stream MetaIO files). It also has a very fast way of parsing directories of DICOM objects to load 3D volumes (by default loading the first series in the directory, if a series ID isn't specified).

I am not certain if ITK provides a way of specifying a collection of DICOM via gzip - I know we have written programs on top of ITK that do that, but interesting idea to add that capability to ITK's imread directly. As you note, it would be easy. It seems outside the scope of this notebook, or am I missing something?

[ericspod]

Line #1.    ngc_api_key = os.environ.get("NGC_API_KEY")

It may be worth noting here that people can just paste their API keys here if they're in an environment that setting an variable isn't feasible.

Reply via ReviewNB

[aylward]

Ok - I am willing to add that, as long as we also note that it is bad practice :). Too often people forget to delete the keys before checking in their code, thereby inadvertently advertising their keys. We shouldn't encourage that practice, imo.

[ericspod]

I had some super minor comments on the notebook but otherwise looks good to me. We can't see the PyVista widgets in the notebook when rendered in Github, I wonder if adding the screenshots in their place as dummies would work. What you could do is run a rendering cell with the code commented and instead displaying an image, then restore the content of the cell with the code before saving the notebook.