Hi everyone !
We preprocessed our data using fMRIprep and now we are using MATLAB/SPM12 for the analysis. I was wondering how to remove from the signal, in matlab, the outliers detected by fMRIprep ? I found how to deal with the motion confounds but not with the outliers 
Thank you for your help,
I wish everyone a very nice day !
Hello,
Assuming you are happy with the outlier threshold default in fMRIPrep (FD of 0.5 or DVARS of 1.5), then you would just need to take the motion_outlierXX confounds from the confounds.tsv file (along with the other nuisance regressors you want to include), and then make a txt file out of those. You can load that into SPM presumably the same way you are doing so with your head motion confounds.
Best,
Steven
I would actually avoid the txt format because it can be hard to figure out what each column is in the models after that.
Save them to a mat file with a names (n X 1 cellstring) and R (n X t array) variables, that you can pass to SPM the same way you would with a txt.
Here a script that should do it and save the output as BIDS like structure.
Note that you will need bids-matlab for this to work: GitHub - bids-standard/bids-matlab: MATLAB / Octave tools for BIDS datasets
% extracts confounds of interest from fmriprep timeseries.tsv
% and saves them for easier ingestion by SPM model specification
path_to_fmriprep = fullfile(pwd, 'fmriprep');
output_folder = fullfile(pwd, 'spm12');
task_label = 'facerepetition';
space_label = 'MNI152NLin2009cAsym';
% set up some regular expression to identify the confounds we want to keep
confounds_of_interest = {'^rot_[xyz]$', '^trans_[xyz]$', '^*outlier*$'};
% index the content of the fmriprep data set and
% figure out which subjects we have
BIDS = bids.layout(path_to_fmriprep, 'use_schema', false);
subjects = bids.query(BIDS, 'subjects');
% prepare the output folder structure
folders = struct('subjects', {subjects}, 'modalities', {{'stats'}});
bids.init(output_folder, 'folders', folders);
for i_sub = 1:numel(subjects)
% create the filter to
filter = struct('sub', subjects{i_sub}, ...
'task', task_label, ...
'desc', 'confounds', ...
'suffix', 'timeseries');
confound_files = bids.query(BIDS, 'data', filter);
% loop through all the desc-confounds_timeseries.tsv
% load it
% get only the condounds we need
% save it in the output folder as a mat file and a TSV
for i = 1:numel(confound_files)
% for mat file
names = {};
R = [];
% for TSV
new_content = struct();
% load
content = bids.util.tsvread(confound_files{i});
confounds_names = fieldnames(content);
% create a logical vector to identify which confounds to keep
% and store confounds to keep in new variables
confounds_to_keep = regexp(confounds_names, ...
strjoin(confounds_of_interest, '|'));
confounds_to_keep = ~cellfun('isempty', confounds_to_keep);
confounds_names = confounds_names(confounds_to_keep);
for j = 1:numel(confounds_names)
% for mat file
names{j} = confounds_names{j};
R(:,j) = content.(confounds_names{j});
% for TSV
new_content.(confounds_names{j}) = content.(confounds_names{j});
end
% save to mat and TSV
output_file_name = bids.File(confound_files{i});
output_file_name.entities.desc = '';
output_file_name.suffix = 'confounds';
output_file_name.path = fullfile(output_folder, ['sub-' subjects{i_sub}], 'stats');
bids.util.tsvwrite(fullfile(output_file_name.path, output_file_name.filename), ...
new_content);
output_file_name.extension = '.mat';
save(fullfile(output_file_name.path, output_file_name.filename), 'names', 'R');
end
end
Tried on this demo dataset : SPM_datasets/spm_facerep_fmriprep - G-Node GIN
To give:
spm12
├── CHANGES
├── dataset_description.json
├── README
└── sub-01
└── stats
├── sub-01_task-facerepetition_confounds.mat
└── sub-01_task-facerepetition_confounds.tsv
will leave a copy of the script here: Extract fmriprep regressors for SPM model specification · GitHub
MANAGED BY INCF