Skip to content

Code for the Medical Segmentation Decathlon Challenge at MICCAI 2018

License

Notifications You must be signed in to change notification settings

ORippler/MSD_2018

Repository files navigation

Medical Segmentation Decathlon Contribution 2018

This repository contains the code for our submission to the MSD 2018. For any questions concerning the code or submission, feel free to open an issue.

The code utilizes the batchgenerators framework maintained by DKFZ, that was modified slightly to work with tf.estimator and tf.data.Dataset and include new features. In detail, spatial_transformations.py and noise_augmentations.py were modified.

It is further in part based on the 3D-Unet implementation from ellisdg.

Prerequisites

Based on the dataset, up to 11 GB of GPU memory may be required. As the data is continuously loaded from Disk, no requirements exist for RAM.

To use the code, please setup a conda environment from the environment.yaml.

Training and predicting a dataset

To train and predict on a challenge task, execute:

  • dataset.py
  • train.py
  • predict.py

in sucession.

About

Code for the Medical Segmentation Decathlon Challenge at MICCAI 2018

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published