This repo is dedicated for our attempt in research using Standford's Musculoskeletal Radiographs (MURA) dataset to detect abnormality

Set up conda environment

bash setup.sh

Activate conda environment

conda activate mura

Train Model

python main.py --env server --mode train

Inference from Model

python main.py --env server --mode predict --run_id <unique identifier> --model_checkpoint <saved model present inside out/<run_id>/checkpoints> --predict_data_dir <image dir for inference>

Back-End: Run the Flask Server

Flask serves the trained model through API, before serving, we need to update the RUN_ID and MODEL_NAME in src/constant.py

python -m src.api

Front-End

Front-End App is developed in React.

To host the front end, change to the dir:

cd src/frontend

Install React Dependency

yarn

Run Front-End

yarn start

Our Trained Model

* Download link for latest model: 
      - https://drive.google.com/drive/folders/1pnME9FFNsNJ8WJcuofn33PbExCN-XzzM?usp=sharing
* Create necessary folder structure: out/<run_id>/checkpoints
* Move the downloaded model inside: out/<run_id>/checkpoints

Citation

@ARTICLE{2017arXiv171206957R,
       author = {{Rajpurkar}, Pranav and {Irvin}, Jeremy and {Bagul}, Aarti and
         {Ding}, Daisy and {Duan}, Tony and {Mehta}, Hershel and {Yang}, Brand
        on and {Zhu}, Kaylie and {Laird}, Dillon and {Ball}, Robyn L. and
         {Langlotz}, Curtis and {Shpanskaya}, Katie and {Lungren}, Matthew P. and
         {Ng}, Andrew Y.},
        title = "{MURA: Large Dataset for Abnormality Detection in Musculoskeletal Radiographs}",
      journal = {arXiv e-prints},
     keywords = {Physics - Medical Physics, Computer Science - Artificial Intelligence},
         year = "2017",
        month = "Dec",
          eid = {arXiv:1712.06957},
        pages = {arXiv:1712.06957},
archivePrefix = {arXiv},
       eprint = {1712.06957},
 primaryClass = {physics.med-ph},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2017arXiv171206957R},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

mahendrathapa/MURA