kailiang-wang/cca

CCA

• Canonical Correlates Analysis (CCA) is a multivariate analysis technique for maximizing the correlations between orthagonalized sets of independent and dependent variates
• The codes provided here are for the CCA as performed in Perry et al., (2017, in review), where
  • CCA linked rs-fMRI patterns to demographic and cognitive measures in 101 cognitively-normal elders
• And are modified from Smith et al's., (2015) HCP investigation

*Adapted from Perry et al., (2017)

Codes included within repository:

• Normalization and decomposition of functional networks
• CCA
• Basic visualization output
• Parcellation templates used in functional network construction

Required dependencies

• MatLab
• PALM
• FSLNETS
• BrainNetViewer (optional for visualisation)

Getting started

1. Data required:

• Functional network matrices of all subjects (i.e. dependent variates in CCA) (connectivity matrices)
• Design Matrix of non-imaging measures (i.e. independent variates in CCA) (DM)
• Motion parameters (i.e. framewise displacement) of functional images (motionFD)
• Centroids of parcellation template employed in functional network construction (COG)

2. Performing the CCA

• Within a MatLab terminal run:
  • [CCAout] = cca_functional(connectivitymatrices, DM, motionFD, COG)

3. Extracting CCA results

• The resultant data will be stored within the Matlab structure CCAout
• Which stores important information in the fields of CCAout, such as:
  • grotU: Individual subjects weights for non-imaging measures captured by each CCA mode
  • grotV: Individual subjects weights for functional connectivity patterns captured by each CCA mode
  • grotR: Correlation between the orthagonalized non-imaging and connectivity patterns
  • conload: Loadings of non-imaging measures onto each modes connectivity patterns (i.e. grotV)
  • grotstats: Parametric statistical output

4. Visualising the CCA output

• Extracted are the connectivity edges and nodes that are most strongly expressed (i.e. top 250 connections) by the first CCA mode
  • For both positive and negative associations with CCA mode
  • Users may want to modify their code, depending on their number of significant CCA modes
• Data is extracted in the .nodes and .edge format required for BrainNet Viewer:
• For example, for the top positive associations:
  • Nodes : CCA_nodes250topposcons_mode1.nodes
  • Edges : CCA_250topposcons_mode1.edge

For any questions and more advanced codes/data please contact Alistair Perry (QIMR Berghofer) (alistairgperry at gmail.com)