Each image location is presumed to be as follows:
wsl_data_dir / data / id.extension
where,

• data is specified as argument during run time itself - currently supports rsna and chexpert
  
• id each unique study or image is accessed by this label - first column of each csv is Id - confirm in wsl_csv_dir
  
• extension is also specified as argument during runtime - it is decoupled from id since each extension requires slightly different loading function
  

wsl_csv_dir / data / file_name.csv
Each csv directory 4 main files - info, train, valid, test
info.csv - first column is Id, the rest are ground truths

• 0/1 for binary classification, simple integers for regression
  
• for binary classification if the 0 and 1 are represent different characterstics, name the column 0_1
  e.g. Male_Female where 0 stands for Male and 1 for Female in the column
  
• for a > 1 classes, make a single column with a list of ground-truth labels for each Id
  
• todo: include support for multi-class regression and multi-class + multi-label ground truths
  train.csv, valid.csv test.csv - single column csv which contains ids to use for current data split
  original.csv - (optional) csv from which info.csv was derived
  

Please refer to wsl / wsl / train.py for how the models are named

Used for storing combined results upon testing of all models - ease of comparison for different architectures and variants

Go to wsl / locations.py Use user variable to reflect your name
Add to the if-else bock to add the location of storage - maintainer recommends using full paths to avoid confusion
The paths added here can be called all over the repo, do not use actual paths anywhere else
TODO: make setting this a part of docker setup

Login to the machine

Make a folder A where your repository will reside - $ mkdir folder-A
Go to foler-A - $ cd folder-A
Inside folder-A - $ git clone https://github.com/mehak-agr/wsl.git
Make a docker - $ sudo docker run --gpus all --ipc=host -it -v /home/mehak.aggarwal/mnt/:/data --name m_wsl projectmonai/monai:latest

Go to folder-A - $ cd /data/2015P002510/...folder-A
Install wsl as a package - $ pip install -e wsl
To train a debug model to ensure things work fine - $ wsl medinet --debug

You can find a comprehensive list of commands and arguments in wsl/main.py
Your trained model will be here - wsl_model_dir / models /

• UNet for segmentation

If you use this code for your research, please cite our paper:

@inproceedings{Medinet2020,
  title={Towards Trainable Saliency Maps in Medical Imaging},
  author={Mehak Aggarwal, Nishanth Arun, Sharut Gupta, Ashwin Vaswani, Bryan Chen, Matthew Li, Ken Chang, Jay Patel, Katherine Hoebel, Mishka Gidwan, Jayashree Kalpathy-Cramer, Praveer Singh},
  booktitle={Machine Learning 4 Health, 2020 Neural Information Processing Systems (NeurIPS)},
  year={2020}
}