CONN Notes

• for resting state data it is perfectly fine to leave the ‘weight’ field to ‘none’: http://www.nitrc.org/forum/message.php?msg_id=5861
• For graph theory:
  
  • ‘correlation coefficient’: the adjacency matrix is formed by selecting the edges with correlation coefficient values above the threshold value that you choose (e.g. if you choose a value of 0.5 then two ROIs are connected if their bivariate correlation is above 0.5).
  • ‘z-score’: the correlation-coefficient values are first transformed to z-scores (i.e. normalized to have zero mean and variance one, separately for each subject), and the adjacency matrix is formed by selecting the edges with z-scores above the chosen threshold value (e.g. if you choose a value of 1 then two ROIs are connected if their correlation coefficients are one standard deviation above the mean -for a given subject).
  • ‘cost’ (the default setting): the adjacency matrix is formed by selecting a fixed percentile (the chosen threshold value) of the edges in each network (those with the largest correlation coefficient values, separately for each subject). Note that this results in graphs for each subject that have the same ‘cost’ (e.g. if you choose a value of 0.15 then each subject graph will have a fixed cost of 0.15, meaning that 15% of all possible edges are present)
  • ‘one-sided’: Only positive correlations are used in graph theory analyses (or more exactly, the sign of the correlations is considered when forming the adjacency matrix based on suprathreshold correlations; this applies not only to cost thresholding but also to the other thresholding methods as well).
    https://www.nitrc.org/forum/message.php?msg_id=6853
• Second-level analyses: beta values represent effect sizes, T-values are the statistics, p-unc represent the false positive level of each individual test, and p-FDR represents the false discovery rate (a correction of the false positive level to account for multiple comparisons). https://www.nitrc.org/forum/message.php?msg_id=6663
• call display 3D function using own graph: conn_mesh_display('', '', '', XYZ, R, .2);
  
  • XYZ is a [N by 3] matrix containing the x/y/z coordinates (in MNI mm) of N ROIs, and R is a [N by N] matrix containing the correlation strengths between your ROIs (set to 0 those connections that you wish not to be displayed)
  • to control the size and color of each ROI sphere, define XYZ to be a structure with fields:
    sph_xyz: [N by 3] matrix of ROI coordinates (xyz mm values)
    sph_r: [N by 1] vector of sphere radius
    sph_c: [N by 3] matrix of sphere colors (rgb 0-1 values)
  • The other input parameters in conn_mesh_display are:
    conn_mesh_display( filenameSURF, filenameVOL, freesurferFolder, XYZ, R, alpha)
    filenameSURF: .img/.nii file containing ‘activation’ values to be displayed at the brain surface (from volume- or surface- based analyses) (e.g. as in the ‘Display Surface’ option in the second-level results explorer)
    filenameVOL: .img/.nii file containing mask to be displayed as ‘activation blobs’ (e.g. as in the ‘Display Volume’ option in the second-level results explorer)
    freesurferFolder: location of surface files used for brain displays (defaults to conn/utils/surf)
    alpha: initial transparency value for brain display (this, as well as many other display options, can be later modified in the GUI)
    https://www.nitrc.org/forum/message.php?msg_id=13041
• NBS: https://www.nitrc.org/forum/message.php?msg_id=21871
• Hierarchical clustering: The hierarchical clustering algorithm in the ‘ROI results explorer’ uses a complete linkage (further distance) method with euclidean distance metric. Distances are computed as a weighted average of differences in connectivity statistics (functional criteria) and differences in spatial location (spatial criteria) between each pair of ROIs. The weighting factor is user-defined (ranging between ‘0’, which uses a purely functional criteria, and ‘1’, which uses a purely spatial criteria). https://www.nitrc.org/forum/message.php?msg_id=9780