Gen7Net - A CNN for classifiying adenocarcinoma of the prostate in MRI images

Gen7Net is a Convolutional Neural Network for classifying prostate cancer in MRI images. It was built with TensorFlow and Keras and is based on MobileNetV2. Training images were taken from the Cancer Imaging Archive.

Contents

shell.nix

Dependency handling was performed with nix-shell for this project. Make sure to have nix installed. Enter the dev environment with:

nix-shell shell.nix

cv.sh

A bash script for converting .dcm to .jpg. Run with:

cv.sh <command> <source> <destination>

Commands

• help | h: display help message
• convert | cv: convert images from .dcm to .jpg
• clean | clear | clr: remove all .jpg files from <source>
• transfer | copy | extract: transfer converted files to <destination> using rsync

Dependencies

Dependencies are handled by running nix-shell against shell.nix.

• detox: rename files and folders with unicode standard
• findutils: find any .dcm or .jpg files in the source folder
• rsync: transfer large quantities of files from A to B
• tree: displays any folder structure in a tree view
• imagemagick: does the heavy lifting of converting between image formats

Gen7Net.ipynb

jupyter-notebook for Gen7Net. I recommend running it with Google Colab. In order to run it locally, I recommend using docker containers with a tensorflow packaging tool like tensorman.

Dependencies

• tensorflow
• keras
• mobilenetv2
• pandas
• numpy
• jupyter-notebook