ardunn / cs188 Goto Github PK

Our project and website for CS 188: Medical Imaging under Fabient Scalzo. Our project focused on applying machine learning to prostate cancer diagnosis.

cs188/              # -> Root folder
├── docs/           # -> Content for our blogposts on our website
└── pyprostate/     # -> Project code
    └── graphics/   # -> MRI images
    └── analysis.py
    └── ensemble.py
    └── first_attempt.py
    └── neural_network.py

Install virtualenv:

$ [sudo] pip install virtualenv

Activate the virtual environment to install the necessary dependencies:

$ source env_188/bin/activate

In ensemble.py, the main function runs our crossvalidate function which executes our machine learning model with specific settings.

ensemble.py takes in the following arguments:

$ python ensemble.py <n> <yes/true/t/y/1 or no/false/f/n/0> 

n specifies the size of the subimages. yes/no specifies if pruning will occur.

Currently, ensemble.py will run the following:

# Lets run an example using T2 MRI, ADC, T2 MRI fitlered, and saving reconstructions!
crossvalidate(model2, filter_on=True, adc_on=True, frequency=0.1, save_reconstruction=True)

On the command line run:

$ python ensemble.py 6 y 

to see our results!

Here are the following keyword arguments you can pass into crossvalidate:

• frequency: The gabor frequency to use if filter_on
• quiet: If True, suppresses intermediate info messages
• silent: If True, suppresses all info messages
• normalized: If True, uses normalized t2 instead of t2 as main dataset.
• filter_on: If True, add a Gabor-filtered dataset of the t2 to the data.
• print_example_vector: If True, prints example vector (ie one point) being fed into the ML algorithm
• save_reconstruction: If True, saves the reconstructed images to .png files.
• adc_on: If True, uses the apparent diffusion constant images as extra features.
• dwi_lvl: If specified, uses the given weight to add diffusion weighted images to the data as extra features.