%% Multiple comparison correction with Threshold-Free Cluster Enhancement
%
% This example demonstrates cosmo_cluster_neighborhood and
% cosmo_montecarlo_cluster_stat
%
% Note: this example shows multiple-comparison for a single subject, but
% the same logic can be applied to a group of subjects to do a group
% analysis.
%
% # For CoSMoMVPA's copyright information and license terms, #
% # see the COPYING file distributed with CoSMoMVPA. #
%% Define data
config=cosmo_config();
data_path=fullfile(config.tutorial_data_path,'ak6','s01');
targets=repmat(1:6,1,10);
chunks=floor(((1:60)-1)/6)+1;
ds = cosmo_fmri_dataset(fullfile(data_path,'glm_T_stats_perrun.nii'),...
'mask',fullfile(data_path, 'brain_mask.nii'), ...
'targets',targets,'chunks',chunks);
% There are 10 chunks, for which the data is assumed to be independent.
% Construct a dataset with 10 samples corresponding to each chunk, with
% the average value across all six targets. Each sample is considered to be
% the same condition, namely the effect of the stimulus-versus-baseline
% effect; thus all target values must be set to 1.
%
% Use either:
% - cosmo_split and cosmo_stack
% - cosmo_average_samples
% - (advanced) cosmo_fx
%
% Assign the result to a variable 'ds_stim'
%%%% >>> Your code here <<< %%%%
%% Define a cluster neighborhood for this dataset and assign the result to
% a variable 'cl_nh'.
% hint: use cosmo_cluster_neighborhood
%%%% >>> Your code here <<< %%%%
% Show a plot with the sorted number of neighbors
% for each voxel
%%%% >>> Your code here <<< %%%%
%% Run cosmo_montecarlo_cluster_stat
% There is one condition per chunk; all targets are set to 1.
% Thus the subsequent anaylsis is a one-sample t-test.
% Note: if this was a group analysis, then each sample (row in ds.samples)
% would contain data from one subject; each unique value in .sa.chunks
% would correspond to one subject; and each unique value in .sa.targets
% would correspond to a condition of interest.
% Since this is a one-sample t-test against a mean of zero, we set this as
% a (required) option
opt=struct();
opt.h0_mean=0;
% set the number of iterations ('niter' option).
% At least 10000 is adviced for publication-quality analyses; because that
% takes quite a while to compute, here we use 200
%%%% >>> Your code here <<< %%%%
% using cosmo_montecarlo_cluster_stat, compute a map with z-scores
% against the null hypothesis of a mean of zero, corrected for multiple
% comparisons. Store the result in a variable named 'tfce_z_ds_stim'
%%%% >>> Your code here <<< %%%%
cosmo_plot_slices(tfce_z_ds_stim);
%% Using the same logic, run a two-sample t-test for primates versus bugs
%%%% >>> Your code here <<< %%%%
cosmo_plot_slices(tfce_z_ds_primate_vs_insects);
