Semi-Supervised Anomaly Segmentation Baseline for Brain MRI

This repository contains the code for "Challenging Current Semi-Supervised Anomaly Segmentation Methods for Brain MRI", accepted to the MICCAI 2021 BrainLes Workshop.

Set-up

Clone the git project:

$ git clone https://github.com/FeliMe/brain_sas_baseline.git

Create a virtual environment and install the requirements:

$ conda create -f environment.yml

Activate the newly created environment:

$ conda activate anomaly_detection

Download ROBEX and SRI ATLAS

Download and install ROBEX from https://www.nitrc.org/projects/robex Download the SRI ATLAS from https://www.nitrc.org/projects/sri24/ and place it into DATAROOT/BrainAtlases/

Download and pre-process Datasets

BraTS

$ python download_data.py --dataset BraTS
$ python download_data.py --dataset BraTS --register

MSLUB

$ python download_data.py --dataset MSLUB
$ python download_data.py --dataset MSLUB --skull_strip
$ python download_data.py --dataset MSLUB --register

WMH

$ python download_data.py --dataset WMH
$ python download_data.py --dataset WMH --skull_strip
$ python download_data.py --dataset WMH --register

MSSEG2015

$ python download_data.py --dataset MSSEG2015
$ python download_data.py --dataset MSSEG2015 --register

Run the experiments (Here Experiment 1 from the paper on BraTS)

$ python baseline.py --test_ds BraTS --img_size 128 --slices_lower_upper 15 125