search

adolphslab / HCP_MRI-behavior

Code for predicting individual differences in behavioral variables (e.g., intelligence, personality) from resting-state fMRI functional connectivity, using data from the Young Adult Human Connectome Project
GNU General Public License v3.0
36 stars 17 forks source link

temporal ICA instead of GSR #2

Open jcrdubois opened 6 years ago

jcrdubois commented 6 years ago

We should implement this temporal ICA approach as an alternative to GSR

paola-g commented 6 years ago

The presented approach follows these steps:

  1. Data are cleaned using ICA-FIX and then MSMAll aligned
  2. A dimensionality reduction step is performed with a group spatial ICA (sICA) with 137 components (probably a method is needed to estimate this dimensionality on new data)
  3. Individual sICA components are created using weighted spatial regression (where weights are a measure of alignment quality)
  4. Temporal ICA is performed on the concatenated individual sICA component time-courses (using FAST-ICA)
  5. The number of temporal components is computed with ICASSO

The code for points 1. to 3. should be part of the HCPpipelines, see especially MSMregression.m.

FAST-ICA and ICASSO are implemented in MATLAB.

So far I haven't found any code that puts everything together.