Starter task in prediction?

Hi all,

I am having my mentee (High school student through ) go over concept-to-clinic code base as a starting point to Deep learning in medical imaging. Thanks for pulling this together! Wondering, if you have any suggestions for a starter task in the prediction part.
We looked at the prediction related issues here( but we could not find a simple task which we can pick up.

Any suggestions are welcome.

Hi @sravya8,

This seems like an ambitious project for a high school student to take on!

No single task in the application jumps out as separable and easy without having a load of experience with the PyData stack and deep learning packages.

My recommendation is that the student take on something concrete but valuable like trying to automatically detect lung orientation from the image (e.g. where is the spinal column) in order to autoflag when one image series is not standard with the orientations of all the others.

The upside here is that they could do the research in a Jupyter notebook (great life skill) directly using the source data without getting bogged down in the existing codebase, and then at the end open a pull request if they have developed some helpful functionality.

Hope that helps!


Hi @isms,

Thanks for your response and your recommendation, I appreciate it!

The main idea of this project is to expose the mentee to the intersection of medical diagnosis and ML. But we are also taking this opportunity to teach life skills of reading others code, leveraging open research and contributing back. And yes, it is pretty ambitious to make further improvements to the prediction part in this time frame.

With respect to your recommendation: Are you suggesting a supervised classification approach? If so, do you know if there exists a training set for it?

In terms of starter tasks in prediction, I do not necessarily mean models. Could be documentation, usability or robustness improvements. Mainly, I would like to help her get her first pull request in - the main purpose is to make a small contribution while learning how to contribute to open source. I was thinking we can add some information about current models and the current metrics to the documentation, which we struggled to find initially. What do you think?

We could also add documentation for radiologists (with your help):

Or fix usability bugs such as:

Let me know your thoughts,