### Selecting components from ICA results

The typical "fingerprint" of an independent component (IC) can be found in BrainVoyager using the "Fingerprint" button on the ICA Overlay dialog. The BOLD fingerprint has high scores on the dimensions: 1. power in band 0.008-0.02Hz 2. power in band 0.02-0.05Hz 3. temporal entropy 4. one lag autocorrelation 5. degree of clustering. When using ICA for artifact removal (task-related motion, etc), other characteristics can be used.

Another way to quickly have an overview of the time courses is to run the script below in Matlab. It calculates the correlation between the time course of the independent components and each condition, and plots the time courses. NeuroElf (previously: BVQXtools) is used in this script to load the time courses of the independent components (*.rtc) and the predictors (*.sdm).

*Figure: First two figures generated by script for independent components (ICs) 1-20 for condition 1 ("faces")*

Script (zipped): component_selection_v01.m

Usage: Load script in Matlab, press 'Run'. If Matlab asks a question about the path, click button 'Add to path'. The script will ask for an *.rtc file containing the time courses of the independent components and for a design matrix file (*.sdm) with predictors. Time courses will be plotted in 10 time courses per condition in one figure.

*Figure: The fingerprint of IC6 suggests it is a BOLD component. The Matlab plot shows how similar the timecourse** of IC6 to the condition "faces" (top of plot) and gives correlation coefficient 0.54.*

### Other ways to sort the components

*Using the Self-Organising group-level ICA plugin*

After running the Self-Organizing Group-Level ICA Plugin via the BrainVoyager "Plugins" menu, the possible BOLD and other clusters can be identified via loading the GRP_SogICA_*ICs.ica file via Analysis > Overlay ICA.

For our example - the RFX ANCOVA data - we figured that IC6 for subject 1 is BOLD. Using the plugin, we now find that for 6 subjects, this component lies in cluster 2, which is probably the BOLD cluster (see figure below).

*Using ICASSO *

The software ICASSO for Matlab can also cluster the ICs into groups like task-related source, vascular source and motion-related source (in the example in Himberg et al, 2004)(although this also requires interpretation). Download the ICASSO software from http://research.ics.aalto.fi/ica/icasso/. This software also requires the FastICA package; the website will show a link to that package. Unzip both packages and add to the Matlab path (for example via addpath(uigetdir()) ). For an example script using functional data from BrainVoyager, see pg. 79 in the Scripting BrainVoyager 20 from Matlab Guide. Please note that the computation for functional data (*.vtc) with 264 time points into 30 components can take longer than one hour.

### References

De Martino F, Gentile F, Esposito F, Balsi M, Di Salle F, Goebel R, Formisano E. (2007). Classification of fMRI independent components using IC-fingerprints and support vector machine classifiers. *Neuroimage*, **34**, 177-194.

Esposito, F., Scarabino, T., Hyvarinen, A., Himberg, J., Formisano, E., Comani, S., Tedeschi, G., Goebel, R., Seifritz, E. & Di Salle, F. (2005). Independent component analysis of fMRI group studies by self-organizing clustering, *Neuroimage*, **25**, 193-205.

J. Himberg, A. Hyvrinen, and F. Esposito (2004) Validating the independent components of neuroimaging time-series via clustering and visualization. *NeuroImage*, **22**(3):1214–1222.

Next > |
---|