Tutorial

ISBI 2018 Tutorial on CXR Anatomy Segmentation

• Intro to Deep Learning with Keras
• Anatomy segmentation with UNet architecture (e.g Keras layers in UNet, modelling, training, testing)
• Data loading, data generators, and image augmentation
• Training/Validation through generator
• Callbacks (e.g. DICE score)
• Visualization

Instructions

Requirements

• A computer with Ubuntu/Mac/Windows operating system.
• High-speed internet access to download packages and repository
• Python 3.6 preferably, although it works with Python 2.7.

IMPORTANT! If you are using Windows machine, you need Python 3.6 since Tensorflow will not be able to install. We strongly recommend using your university's server or an AWS server with Linux instead. It comes with 750 hours of free computing: https://aws.amazon.com/ec2/

• VirtualEnv (see Linux/Mac instructions: https://virtualenv.pypa.io/en/stable/installation/, see example Windows instructions: http://timmyreilly.azurewebsites.net/python-pip-virtualenv-installation-on-windows/)

Installation Steps

1. Create a folder e.g. ~/Desktop/Tutorial/
2. Download requirements.txt file from https://ibm.box.com/s/e8yhc80kqkfvfz6mbfnbwvcypnkz3944 and save it into the previous folder e.g. ~/Desktop/Tutorial/
3. Create a VirtualEnv (e.g. ISBI-Tutorial-Environment) . To learn how to create a VirtualEnv and understand more here: https://virtualenv.pypa.io/en/stable/userguide/#usage
4. Activate your new VirtualEnv. To learn how to activate see more here: https://virtualenv.pypa.io/en/stable/userguide/#activate-script
5. Now that you are inside your virtual environment go to your folder e.g. cd ~/Desktop/Tutorial/
6. From the terminal run pip3 install -r requirements.txt if your VirtualEnv is running Python3 or pip install -r requirements.txt for Python2 to install the libraries. That might take a few minutes.
7. Once finished, make sure Keras/Tensorflow work (see figure below) by running python3 for Python3 users (python for Python2 users ) and then inside the interpreter run import keras. See figure below.

8. Type exit() to exit the python interpreter.

Post-Installation Steps

1. Download this project from Github (https://github.com/alexkarargyris/ISBI_2018_Tutorial/archive/master.zip) and unzip its files inside your folder e.g. ~/Desktop/Tutorial/
2. Download the pretrained network with 10x10 kernels from here https://ibm.box.com/s/ane1j8h4g2ra1rookny1jz7gxwnjodxw and place it inside your folder ~/Desktop/Tutorial/
3. Download the pretrained network with 2x2 kernels from here https://ibm.box.com/s/bh4dik59b3h37t184eidmtnmuxt3a2o8 and place it inside your folder ~/Desktop/Tutorial/
4. With your Virtual Environment activated navigate to your folder e.g. ~/Desktop/Tutorial and execute jupyter notebook . The browser should open to following screen.

---

About the data

Disclaimer: Please do not share the data set outside of your research group/organization, but forward new data set requests to Sameer Antani, project leader at National Library of Medicine of the National Institutes of Health. *

IRB Details: The dataset were de-identified by the data providers and were exempted from IRB review at their institutions. The data set use and public release were exempted from IRB review (No. 5357) by the NIH Office of Human Research Projections Programs.

References

1.  Candemir S, Jaeger S, Musco J, Xue Z, Karargyris A, Antani SK, Thoma GR, Palaniappan K. Lung segmentation in chest radiographs using anatomical atlases with nonrigid registration. IEEE Trans Med Imaging. 2014 Feb;33(2):577-90. doi: 10.1109/TMI.2013.2290491. PMID: 24239990
2.  Jaeger S, Karargyris A, Candemir S, Folio L, Siegelman J, Callaghan FM, Xue Z, Palaniappan K, Singh RK, Antani SK. Automatic tuberculosis screening using chest radiographs. IEEE Trans Med Imaging. 2014 Feb;33(2):233-45. doi: 10.1109/TMI.2013.2284099. PMID: 24108713

*https://ceb.nlm.nih.gov/repositories/tuberculosis-chest-x-ray-image-data-sets/