GMM_MRI

Code repository for a Gaussian Mixture Model to segment MRI data. This repository is for Statistical Learning III at Portland State University.

Data is omitted. Download available data into the data folder. Expected files are in data.txt and gt.txt.

Prerequisites

• Python 3.0
• TensorFlow
• Scipy
• Numpy
• Matplotlib

Directory structure

data

Data folder that contains all avaliable images with list of filenames.

latex

Latex source file for derivation.pdf

Source Files

data.py

Files for reading in images and parsing out ground truth into numpy arrays

kmeans.py

Applies kmeans to data using TensorFlow. Outputs an .npy of cluster means into data folder.

semi_gmm.py

Applies semi-supervised GMM using EM, with test images as missing data.

sup_gmm.py

Applies supervised GMM using EM with only training images, and applies E step on test images

unsup_gmm.py

Applies unsupervised GMM using EM with all available data

Documents

derivation.pdf

Derivation of update rules for semi-supervised GMM using EM

slides.pdf

Slides for output images comparing kmeans, unsupervised, supervised, and semi-supervised GMM on test image