Leonardo Gutiérrez-Gómez, Jakub Vohryzek, Benjamin Chiêm, Philipp S. Baumann, Philippe Conus, Kim Do Cuenod, Patric Hagmann and Jean-Charles Delvenne.
Here you can find the scripts to reproduce the results of our paper. Check out the preprint here
This code was tested on Linux/Mint 18.2 with python 2.7.
The entire dataset can be downloded from Zenodo. The Funcitonal and Structural connectivity matrices should be vectorized and saved in the folders FC and SC.
The following script will run recursive feature elimination, support vector machines (RFE-SVM) to select features, compute the stability of selected biomarkers and predicting schizophrenia.
Example,
python mainScript_abs.py -connectivity Structural -resolution 83
Will run RFE-SVM on structural connectomes with 83 x 83 parcellations. The results are saved in the mat/abs_subcortical folder.
Plotting accuracy and stability for different parameters. Figures will be saved in the results/ folder as .png files.
python plotting_accuracy_stability.py
example, structural 83 x 83 connectome:
You can also plot the selected biomarkers and visualize them as a 'connectivity' matrices:
python plotting_selected_features_matrices.py
Best tradeoff between accuracy and stability.
python plotting_stab_vs_acc.py
python acore_distribution.py -connectivity Multimodal -resolution 83 -type Structural
This script will compute the a-core of the structural biomarkers extracted from the Multimodal connectome of parcellation 83 x 83.
Soon bibtex entry.
Please send any questions you might have about the code and/or the paper to leonardo.gutierrez@list.lu, jakub.vohryzek@queens.ox.ac.uk or benjamin.chiem@uclouvain.be.