/pocketnet

Code for "PocketNet: A Smaller Neural Network for Medical Image Analysis."

Primary LanguageJupyter NotebookMIT LicenseMIT

PocketNet: A Smaller Neural Network For Medical Image Analysis

Code and materials for:

Celaya, A., Actor, J. A., Muthusivarajan, R., Gates, E., Chung, C., Schellingerhout, D., Riviere, B., Fuentes, D. PocketNet: A Smaller Neural Network for 3D Medical Image Segmentation. In Medical Image Computing and Computer Assisted Intervention -- MICCAI 2021 (2021). Submitted.

https://arxiv.org/abs/2104.10745

Data

MICCAI Liver and Tumor Segmentation Challenge 2017 dataset (LiTS) - https://competitions.codalab.org/competitions/17094#learn_the_details-overview

Neurofeedback Skull-stripped repository (NFBS) - http://preprocessed-connectomes-project.org/NFB_skullstripped/

MICCAI Brain Tumor Segmentation Challenge 2020 dataset (BraTS) - https://www.med.upenn.edu/cbica/brats2020/registration.html

COVIDx8B dataset - https://github.com/lindawangg/COVID-Net/blob/master/docs/COVIDx.md

Usage instructions

  1. Download each dataset and save to a directory of your choice.
  2. Preprocess the data with preprocess.ipynb
  3. Train models using train_pocketnet.ipynb
  4. Train for model saturation using saturation.ipynb
  5. Run performance profiling with performance_profile.ipynb

A generic implementation of each architecture (pocket and non-pocket versions) is available in pocketnet.ipynb.

Convert ipynb to py

ipython3 nbconvert preprocess.ipynb --to python