Name: cluster-glm
Owner: Kimel Family Translational Imaging-Genetics Research Lab
Description: work done on hawco et al. 2017
Forked from: josephdviviano/cluster-glm
Created: 2017-10-06 00:09:54.0
Updated: 2017-10-06 00:10:00.0
Pushed: 2017-01-18 23:42:08.0
Homepage: null
Size: 223696
Language: Matlab
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|
Other Committers
| User | Most Recent Commit | # Commits |
|---|
cluster_gridsearch.m
Uses a clustering-stability method to termine the model order (number of clusters) to be used when partioning the data (spoiler: 2 or 3 is the correct answer).
This also could be used to compare distance functions, but that has little influence on the output. This does not do a good job at comparing linkage functions, as our data has some extreme outliers. Therefore, complete, single, average, centroid clustering all produce clusters with single individuals, and a second cluster with the remainder of the data. This produces extremely stable, but uninstering, solutions.
Also, when subdividing the 'oddball' group, the clusters become quite small, as the individuals are quite different from one another.
Outputs cluster_grid_search.mat, and cluster_stability.fig.
export.m
Takes the settings found in cluster_gridsearch.m, and runs a GLM on each cluster to test whether each voxel is has a mean value > 0. The outputs of these GLM runs are exported a .nii files.
Also plots a dendrogram + distance matrix of the selected cluster settings.
Outputs cluster_final.mat and stat_hist.fig.
project-to-surface.sh
Relies on the connectome-workbench. Not actually functional, but a reminder of how to convert the output .nii files from export.m to dscalar.nii files that can be opened in connectome workbench.