Usage: 3dECM [options] dset
Computes voxelwise eigenvector centrality (ECM) and
stores the result in a new 3D bucket dataset as floats to
preserve their values. ECM of a voxel reflects the strength
and extent of a voxel's global connectivity as well as the
importance of the voxels that it is directly connected to.
Conceptually the process involves:
1. Calculating the correlation between voxel time series for
every pair of voxels in the brain (as determined by masking)
2. Calculate the eigenvector corresponding to the largest
eigenvalue of the similarity matrix.
Guaranteeing that the largest eigenvector is unique and therefore,
that an ECM solution exists, requires that the similarity matrix
is strictly positive. This is enforced by either adding one to
the correlations as in (Lohmann et. al. 2010), or by adding one
and dividing by two (Wink et al. 2012).
Calculating the first eigenvector of a whole-brain similarity matrix
requires a lot of system memory and time. 3dECM uses the optimizations
described in (Wink et al 2012) to improve performance. It additionally
provides a mechanism for limited the amount of system memory used to
avoid memory related crashes.
The performance can also be improved by reducing the number of
connections in the similarity matrix using either a correlation
or sparsity threshold. The correlation threshold simply removes
all connections with a correlation less than the threshold. The
sparsity threshold is a percentage and reflects the fraction of
the strongest connections that should be retained for analysis.
Sparsity thresholding uses a histogram approach to 'learn' a
correlation threshold that would result in the desired level
of sparsity. Due to ties and virtual ties due to poor precision
for differentiating connections, the desired level of sparsity
will not be met exactly, 3dECM will retain more connections than
requested.
Whole brain ECM results in very small voxel values and small
differences between cortical areas. Reducing the number of
connections in the analysis improves the voxel values and
provides greater contrast between cortical areas
. Lohmann G, Margulies DS, Horstmann A, Pleger B, Lepsien J, et al.
(2010) Eigenvector Centrality Mapping for Analyzing
Connectivity Patterns in fMRI Data of the Human Brain. PLoS
ONE 5(4): e10232. doi: 10.1371/journal.pone.0010232
Wink, A. M., de Munck, J. C., van der Werf, Y. D., van den Heuvel,
O. A., & Barkhof, F. (2012). Fast Eigenvector Centrality
Mapping of Voxel-Wise Connectivity in Functional Magnetic
Resonance Imaging: Implementation, Validation, and
Interpretation. Brain Connectivity, 2(5), 265-274.
doi:10.1089/brain.2012.0087
Options:
-full = uses the full power method (Lohmann et. al. 2010).
Enables the use of thresholding and calculating
thresholded centrality. Uses sparse array to reduce
memory requirement. Automatically selected if
-thresh, or -sparsity are used.
-fecm = uses a shortcut that substantially speeds up
computation, but is less flexibile in what can be
done the similarity matrix. i.e. does not allow
thresholding correlation coefficients. based on
fast eigenvector centrality mapping (Wink et. al
2012). Default when -thresh, or -sparsity
are NOT used.
-thresh r = exclude connections with correlation < r. cannot be
used with FECM
-sparsity p = only include the top p% (0 < p <= 100) connectoins in the calculation
cannot be used with FECM method. (default)
-do_binary = perform the ECM calculation on a binarized version of the
connectivity matrix, this requires a connnectivity or
sparsity threshold.
-shift s = value that should be added to correlation coeffs to
enforce non-negativity, s >= 0. [default = 0.0, unless
-fecm is specified in which case the default is 1.0
(e.g. Wink et al 2012)].
-scale x = value that correlation coeffs should be multiplied by
after shifting, x >= 0 [default = 1.0, unless -fecm is
specified in which case the default is 0.5 (e.g. Wink et
al 2012)].
-eps p = sets the stopping criterion for the power iteration
l2|v_old - v_new| < eps*|v_old|. default = .001 (0.1%)
-max_iter i = sets the maximum number of iterations to use in
in the power iteration. default = 1000
-polort m = Remove polynomial trend of order 'm', for m=0..3.
[default is m=1; removal is by least squares].
Using m=0 means that just the mean is removed.
-autoclip = Clip off low-intensity regions in the dataset,
-automask = so that the correlation is only computed between
high-intensity (presumably brain) voxels. The
mask is determined the same way that 3dAutomask works.
-mask mmm = Mask to define 'in-brain' voxels. Reducing the number
the number of voxels included in the calculation will
significantly speedup the calculation. Consider using
a mask to constrain the calculations to the grey matter
rather than the whole brain. This is also preferable
to using -autoclip or -automask.
-prefix p = Save output into dataset with prefix 'p'
[default prefix is 'ecm'].
-memory G = Calculating eignevector centrality can consume a lot
of memory. If unchecked this can crash a computer
or cause it to hang. If the memory hits this limit
the tool will error out, rather than affecting the
system [default is 2G].
Notes:
* The output dataset is a bucket type of floats.
* The program prints out an estimate of its memory used
when it ends. It also prints out a progress 'meter'
to keep you pacified.
-- RWCox - 31 Jan 2002 and 16 Jul 2010
-- Cameron Craddock - 13 Nov 2015
-- Daniel Clark - 14 March 2016
=========================================================================
* This binary version of 3dECM is compiled using OpenMP, a semi-
automatic parallelizer software toolkit, which splits the work across
multiple CPUs/cores on the same shared memory computer.
* OpenMP is NOT like MPI -- it does not work with CPUs connected only
by a network (e.g., OpenMP doesn't work across cluster nodes).
* For some implementation and compilation details, please see
https://afni.nimh.nih.gov/pub/dist/doc/misc/OpenMP.html
* The number of CPU threads used will default to the maximum number on
your system. You can control this value by setting environment variable
OMP_NUM_THREADS to some smaller value (including 1).
* Un-setting OMP_NUM_THREADS resets OpenMP back to its default state of
using all CPUs available.
++ However, on some systems, it seems to be necessary to set variable
OMP_NUM_THREADS explicitly, or you only get one CPU.
++ On other systems with many CPUS, you probably want to limit the CPU
count, since using more than (say) 16 threads is probably useless.
* You must set OMP_NUM_THREADS in the shell BEFORE running the program,
since OpenMP queries this variable BEFORE the program actually starts.
++ You can't usefully set this variable in your ~/.afnirc file or on the
command line with the '-D' option.
* How many threads are useful? That varies with the program, and how well
it was coded. You'll have to experiment on your own systems!
* The number of CPUs on this particular computer system is ...... 1.
* The maximum number of CPUs that will be used is now set to .... 1.
=========================================================================
++ Compile date = Oct 17 2024 {AFNI_24.3.03:linux_ubuntu_24_64}
