Crossvalidation searchlight

Contents

Define data

config = cosmo_config();
data_path = fullfile(config.tutorial_data_path, 'ak6', 's01');

% In this exercise, output (NIFTI files) will be written to this directory
output_data_path = config.output_data_path;

data_fn = fullfile(data_path, 'glm_T_stats_perrun.nii');
mask_fn = fullfile(data_path, 'brain_mask.nii');
ds = cosmo_fmri_dataset(data_fn, 'mask', mask_fn, ...
                        'targets', repmat(1:6, 1, 10), ...
                        'chunks', floor(((1:60) - 1) / 6) + 1);

% remove constant features (due to liberal masking)
ds = cosmo_remove_useless_data(ds);

Neighborhood definition

% Use a searchlight with a more-or-less constant number of voxels,
% both near the edges of the brain and in the center of the brain.
nvoxels_per_searchlight = 100;

% Define a spherical neighborhood with approximately
% 100 voxels around each voxel using cosmo_spherical_neighborhood,
% and assign the result to a variable named 'nbrhood'
% >@@>
nbrhood = cosmo_spherical_neighborhood(ds, 'count', nvoxels_per_searchlight);
% <@@<

+00:00:01 [####################] -00:00:00  mean size 99.8

Define the measure as

define the measure as cosmo_crossvalidation_measure >@@>

measure = @cosmo_crossvalidation_measure;
% <@@<

% Define measure arguments.
% Ideally we would like to use nfold_partitioner (and you probably
% want to do this for a publication-quality), but this takes quite long.
% Instead, we take a short-cut here and use an
% odd-even partitioning scheme that has only one fold (test on odd chunks,
% test on even) using cosmo_oddeven_partitioner.

measure_args = struct();

% As the 'partitions' argument, use an odd-even partitioner
% >@@>
measure_args.partitions = cosmo_oddeven_partitioner(ds, 'half');
% <@@<

% Use cosmo_classify_lda as classifier argument for the measure
% >@@>
measure_args.classifier = @cosmo_classify_lda;
% <@@<

Run searchlight

% Using the neighborhood, the measure, and the measure's arguments, run the
% searchlight using cosmo_searchlight. Assign the result to a variable
% named 'ds_cfy'

% >@@>
ds_cfy = cosmo_searchlight(ds, nbrhood, measure, measure_args);
% <@@<

+00:00:10 [####################] -00:00:00  

Visualize and store the results in a NIFTI file

% Visualize the results using cosmo_plot_slices
cosmo_plot_slices(ds_cfy);

% Set output filename
output_fn = fullfile(output_data_path, ...
                     sprintf('lda_odd-even_accuracy_d%.0fvx.nii', nvoxels_per_searchlight));

% Write output to a NIFTI file using cosmo_map2fmri
% >@@>
cosmo_map2fmri(ds_cfy, output_fn);
% <@@<

Published with MATLAB® R2024a