Group ICA of resting-state data: a comparison
Veronika Schöpf1, Christian Windischberger, Christian H Kasess
1MR Centre of Excellence, Medical University Vienna, Vienna, Austria. veronika.schoepf@meduniwien.ac.at
Magma (New York, N.Y.)
|June 4, 2010
Summary
Independent Component Analysis (ICA) for group functional magnetic resonance imaging (fMRI) was compared using GIFT and PICA software. Both methods adequately identify resting-state networks (RSNs), with GIFT yielding more RSNs at lower component levels.
Area of Science:
- Neuroimaging
- Computational Neuroscience
Background:
- Independent Component Analysis (ICA) is widely used for fMRI data, but group ICA methods vary.
- Two prominent group ICA software packages are GIFT and PICA, both employing spatial ICA for group map estimation.
Purpose of the Study:
- To compare the performance of GIFT and PICA for group Independent Component Analysis (ICA) in resting-state fMRI.
- To assess the number of resting-state networks (RSNs) detected and computational load for each method.
Main Methods:
- Applied GIFT and PICA to resting-state fMRI data from 28 healthy subjects.
- Evaluated performance across component levels (5-35) using default implementations.
- Assessed the number of detected RSNs and computational efficiency.
Main Results:
- GIFT identified more RSNs than PICA at lower component estimation levels.
- At individually determined component levels, both GIFT and PICA detected comparable RSNs.
- Spatial and temporal comparisons showed no significant differences in detected RSNs between the methods, despite some map variability.
Conclusions:
- Both GIFT and PICA are adequate for group ICA in resting-state fMRI, yielding comparable RSNs.
- The primary difference between GIFT and PICA lies in their calculation times.
