Stanford’s MURA Bone X-Ray Deep Learning Competition

Raise: TODO

The goal of this project is to …
Raise: TODO

The instructions below will get you a copy of the project up and running on your local machine for development and testing purposes.

Table of Contents

1. Data
2. Prerequisites
3. Steps
4. Results
5. Authors

Data

Data is openly availble from Stanford’s ML Lab: https://stanfordmlgroup.github.io/competitions/mura/

Here are two samples of negative and positive data:

Some high level EDA findings:

1. asd
2. ads
3. ads

Prerequisites

A list of conda/pip environment dependencies can be found in the environments.yml file. To create a conda env with all of the dependencies run the create_conda_env.sh shell script. We are also using Tensorflow and Keras with GPU support.

Steps

1. Download the MURA dataset and unzip it into a a location of your chosing.
2. Run the shell script env_setup.sh This will create the conda environment that we used to build the model.
3. Run the shell script create_ini_files.sh This will create a config.ini file where you will need to put a path to your data. for example my path is: /Users/keil/datasets/mura/
4. Run merge_csv.py to create the merged sample and full csv files. two csvs will be created in the sample_data/ directory and two csvs in your MURA data path location.
5. Run the data_pipeline.py file and congratz you are where we are! …
6. more to come…

Results

Raise: TODO

Authors

• Kyle Shannon
• Chris Chen
• Laura Wilke