Data cleaning, analysis, and visualization code for the manuscript at https://openreview.net/forum?id=Hkg0j9sA1V

Instructions

In order to reproduce the figures from the submitted manuscript, the following steps are required:

• Create a python environment containing PyTorch, lagomorph, pandas, and jupyter
• Download the OASIS data and convert to HDF5
• Run the script run_all.py in the current directory
• Run the notebook OASISPlots.ipynb

Below are details for each of these steps.

NOTE: At this time, lagomorph only supports CUDA devices, so you must have an nvidia CUDA gpu available when running the analysis script.

Creating a python environment

Using either conda or python's venv module, create a new virtualenv, activate it, and run the following:

pip install pytorch lagomorph pandas jupyter

Downloading the data

First, sign up for data access for the OASIS-3 project and agree to the terms and conditions at https://oasis-brains.org/

Once you have a username and password for central.xnat.org, run the following commands in the data directory:

XNATUSER=<fill in your xnat username>
./download_oasis_skullstripped.sh ./all_freesurfers.csv skullstripped $XNATUSER
./create_oasis3_h5.py

Run analysis

With the virtual environment active, run

python3 run_all.py

This will create many results files in the current directory with extensions .pth and .h5. If you re-run an interrupted run, the analysis should find these and load them instead of recomputing the intermediate files. Clear them all out if you'd like to play with parameters and re-run the analysis from scratch.

Visualize results

Once the pth and h5 files are computed by the previous step, the visualizations can be generated. Simply load the OASISPlots.ipynb notebook in jupyter and rerun it to recreate the visualizations in the manuscript.