This repository contains the code that I have written to assess the correlation
between age and brain volumetric measurements in 20 MRI images.
Implementation of a groupwise registration pipeline to create a groupwise space of 10
previously segmented images. This registration task was implemented using NiftyReg
tools (see http://cmictig.cs.ucl.ac.uk/wiki/index.php/NiftyReg_documentation and
http://cmictig.cs.ucl.ac.uk/wiki/index.php/NiftyReg_install). This groupwise space
was then used to generate mean tissue probability maps for non brain tissue,
cerebrospinal fluid (CSF), white matter (WM) and grey matter (GM) for this population.
The tissue probability maps were then propagated into the space of 20 unsegmented
images, which were used as a priori information for their segmentation. A Gaussian
Mixture Model (GMM) was implemented to segment these images, optimised through an
Expectation-Maximisation scheme. A Markov random field was embedded into the
segmentation framework to introduce a spatial smoothness in the label estimation
process. See "Automated Model-Based Tissue Classification of MR Images of the Brain"
in the 'Papers' folder.
As MRI acquisition usually suffers from magnetic field intensity non-uniformity,
the robustness of the GMM framework was improved by adding a bias field correction
component to the probabilistic model. See "Automated Model-Based Bias Field Correction
of MR Images of the Brain" in the 'Papers' folder.
Optimisation of segmentation implementation parameters was achieved by using a DICE
image similarity metric to compare test segmentations to a ground truth segmentation.
Statistical analysis was then performed to assess the relationship between brain
volume (having normalised by total intracranial volume) and age, as well as the
relationship between GM/WM and age in the 20 segmented MRI images.
Please refer to Written_Report.pdf for a complete description of this project.