AFNI Interactive Clusterizing

AFNI now has an interactive cluster-izing interface, that works with the thresholding. The method is the same as used in program 3dclust. The two slides below show the interface and some of its results. For the true and fearless AFNI acolyte, these should be enough to get started.

For the less true and more fearful, some of the capabilities:

• To get started, click on 'Clusterize' on the AFNI control panel, select rmm and vmul from the chooser that pops up, then press 'Apply' or 'Set'.
  • rmm=0 means nearest neighbor clustering, in which case vmul is the minimum allowable voxel count in a cluster.
    • Otherwise, rmm and vmul are the cluster connectivity radius (in mm) and minimum volume size (in microliters), as in program 3dclust
  • You will probably want to adjust the threshold slider appropriately.
  • Click on the 'Rpt' button to get the cluster report table (only works when 'See Overlay' is turned on and at least one image viewer window is open — since otherwise, no clustering will actually be done since there is no need to display the overlay).
  • Click on the 'Clear' button (below Clusterize) to turn off the interactive clustering so you can see all above-threshold voxels again.
    • Clear will also close the report panel window.
    • A '*' character appears on the Clusterize button when interactive clustering is active, and on the Clear button when clustering is turned off.


• After thresholding, AFNI finds clusters of voxels as defined by the rmm parameter, then removes clusters below the vmul size limit from the color overlay.
  • Clustering is done in the voxel space of the overlay dataset, not as interpolated to the voxel space of the underlay.
  • The top of the report panel shows the number of voxels that are left and the number of those that were removed.
  • The cluster report panel changes interactively when you change the threshold slider.
  • The '#' label just to the left of '|xyz' shows the cluster index in which the image viewer crosshair focus currently lies. If this label shows '??', then the crosshairs are not in one of the reported clusters.
  • N.B.: The popup menu above the AFNI threshold slider lets you control if all values, only positive values, or only negative values pass the threshold magnitude test ("all values" is the default). The interactive clusterizing also obeys these options.
  • There are no interactive clustering options equivalent to the following 3dclust options:
      -noabs   -isovalue   -isomerge


• The '3dclust' button writes (to the text window in which you started AFNI) the 3dclust command equivalent to the current interactive clusterize setup.


• The 'Save Table' button writes the current report table to a 1D file, in a format similar to that produced by the 3dclust program.
  • '# line comments at the top of the output 1D file indicate the cluster setup used to produce this output.


• The text entry field to the right of 'Save Table' sets the prefix for all output files.
  • e.g., Clust_table.1D, Clust_01_mean.1D, et cetera.


• The report table shows number of voxels and the peak ('Peak') or center-of-mass ('Cmass') coordinates of each cluster (a maximum of 15 clusters will be shown in this report).
  • Coordinate order (RAI or LPI) is same as in main AFNI controller window.
  • 'Jump' will move the crosshair focus to the given location (cluster peak location or center-of-mass, each as weighted by the current OLay sub-brick).
  • 'Flash' will turn the color overlay for a given cluster off and on rapidly a few times, to help in identifying where that cluster is.


• You can plot or save data extracted using a cluster's voxels as a mask.
  • The dataset from which the data is extracted is chosen with the 'Aux Dataset' button; this dataset must have the same voxel grid as the current overlay dataset (which is the one from which the clusters are defined).
    • If you switch the AFNI overlay dataset to one that has a different voxel grid, then the Aux Dataset setting will be cleared in the report panel.
  • 'From' and 'To' specify the range of sub-bricks to be processed from the auxiliary dataset.
    • If the To setting is past the end of the dataset, that will be interpreted to mean that you want To to be the last sub-brick in the dataset; thus, the default values From=0 and To=99999 mean to process all the sub-bricks in Aux Dataset — unless you more than 100,000 sub-bricks in your dataset!
  • You can extract the 'Mean' or 'PC#1' (first principal component) time series from the auxiliary dataset, computed from all the voxels in a given cluster, over the given sub-brick range.
  • The 'Hist' option instead computes a histogram of all the values extracted from the auxiliary dataset.
  • A right-click popup menu on the top 'Voxels survived ...' label (the top of the report panel) lets you set the range of values that are used in the histogram; by default, all values will be used, but you can specify a minimum and maximum.
  • 'Plot' displays a graph of the extracted results (a timeseries or a histogram plot) to the screen.
  • 'Save' saves the extracted results to a 1D file, for your further processing pleasure.

RW Cox — 28 Dec 2007