Diffusion Research Scripts (Thomas Bolton) (master) Browse

SlidingWindow.m
bdfe7aadccccmaster

SlidingWindow.m
View Options

			%% This function performs sliding window analysis on the data of one
			% subject. It can perform it the standard way, or the 'IFSC' way
			%
			% Inputs:
			% - X is the data of all the subjects that we want to consider (cell array with each cell n_ROI x n_TP)
			% - W is the window size (in samples)
			% - Step is the window Step (in samples)
			% - ConnType is the type of measure to use (e.g. 'Pearson')
			% - SWType is the type of sliding window correlation to perform
			%
			% Outputs:
			% - Y is the output (n_pairs x n_swtp)
			function [Y,CM_stat] = SlidingWindow(X,W,Step,ConnType,SWType)

			% Number of time points of the data
			n_TP = size(X{1},2);

			% Number of regions
			n_ROI = size(X{1},1);

			% Number of subjects
			n_subjects = length(X);

			% Number of pairs of regions
			n_pairs = n_ROI * (n_ROI-1) / 2;

			% Current end of the window; we want to iterate the computations until
			% the window end is past the end of the data
			WindowEnd = W;
			WindowStart = 1;

			time_w = 1;

			full_index = 1;

			Y = cell(1,n_subjects);
			CM_stat = cell(1,n_subjects);

			while WindowEnd <= n_TP

			%disp(['Currently in round ',num2str(full_index),' of computations']);

			switch SWType
			% Normal FC across regions from the same subject
			case 'FC'

			for i = 1:n_subjects

			% Samples the part of the time course of interest
			tmp = X{i}(:,WindowStart:WindowEnd);

			% Computes correlation
			switch ConnType
			case 'Pearson'
			% C is a n_ROI x n_ROI matrix
			C = corr(tmp');

			% Gets a vector with values
			Y{i}(:,time_w) = jUpperTriMatToVec(C);

			otherwise
			errordlg('Not yet implemented !');
			end

			clear tmp
			end

			% FC across the regions from different subjects
			case 'ISFC'

			% For each subject, we want to compute the estimate for a
			% given region pair comparing this particular region to all
			% the other subject's paired area
			for i = 1:n_subjects

			% tmp2 will contain our samples of time courses for all
			% subjects different from subject 1
			tmp2 = nan(n_ROI,W,n_subjects-1);

			% tmp is our data for subject i
			tmp = X{i}(:,WindowStart:WindowEnd);

			% Takes the indices of all the other subjects
			s_idx = 1:n_subjects;
			s_idx(i) = [];

			% Loops through those, and samples the data for them
			% into tmp2
			for j = 1:n_subjects-1
			tmp2(:,:,j) = X{s_idx(j)}(:,WindowStart:WindowEnd);
			end

			% Uses the right connectivity measure, and computes the
			% correlation between the mean time courses of all
			% subjects (except subject i) and subject i time
			% courses
			switch ConnType
			case 'Pearson'

			% I make my matrix symmetric
			tmp_corr = corr(tmp',squeeze(mean(tmp2,3))');
			tmp_corr_sym = (tmp_corr + tmp_corr')/2;

			Y{i}(:,time_w) = jUpperTriMatToVec(tmp_corr_sym);

			otherwise
			errordlg('Not yet implemented !');
			end

			clear tmp2
			clear tmp
			end

			% Here, we want to implement the bootstrapping option from Chen
			% et al. 2016 (NI) for a better ISFC computation
			case 'Bootstrap'

			% Number of bootstrapping folds
			n_boot = 100;

			% We will consider each subject one after the other...
			for i = 1:n_subjects

			disp(['Subject ',num2str(i),'...']);

			% This will contain the correlation coefficients
			% computed between subject i and all possible other
			% subjects
			CM_boot = zeros(n_ROI,n_ROI,n_boot);

			% We want to run the same assessment many many times
			for boot = 1:n_boot

			tmp_corr_avg = zeros(n_ROI,n_ROI);

			% Samples the indices of the subjects to keep (with
			% replacement)
			idx_boot = randsample(n_subjects-1,n_subjects-1,true);

			% Takes the indices of all the other subjects
			s_idx = 1:n_subjects;
			s_idx(i) = [];

			% We will compute ISFC with every possible second
			% subject as selected from idx_boot
			for j = 1:length(idx_boot)

			% For the considered pair, n_ROI x n_ROI matrix of
			% correlation
			tmp_corr = corr(X{i}(:,WindowStart:WindowEnd)',X{s_idx(j)}(:,WindowStart:WindowEnd)');

			% I make my matrix symmetric
			tmp_corr_avg = tmp_corr_avg + ((tmp_corr + tmp_corr')/2);
			end

			tmp_corr_avg = tmp_corr_avg/length(idx_boot);

			CM_boot(:,:,boot) = tmp_corr_avg;
			end

			% This contains the data of interest
			Y{i}(:,time_w) = jUpperTriMatToVec(median(CM_boot,3));
			end
			end

			% We shift the window by the step size
			WindowStart = WindowStart + Step;
			WindowEnd = WindowEnd + Step;
			time_w = time_w + 1;
			full_index = full_index + 1;
			end

			switch SWType
			case 'ISFC'
			for i = 1:n_subjects
			tmp2 = nan(n_ROI,n_TP,n_subjects-1);

			tmp = X{i};

			s_idx = 1:n_subjects;
			s_idx(i) = [];

			for j = 1:n_subjects-1
			tmp2(:,:,j) = X{s_idx(j)};
			end

			switch ConnType
			case 'Pearson'
			tmp_corr = corr(tmp',squeeze(mean(tmp2,3))');
			tmp_corr_sym = (tmp_corr + tmp_corr')/2;
			CM_stat{i} = tmp_corr_sym;
			otherwise
			errordlg('Not yet implemented !');
			end
			end

			case 'FC'
			for i = 1:n_subjects

			tmp = X{i};

			switch ConnType
			case 'Pearson'
			CM_stat{i} = corr(tmp');
			otherwise
			errordlg('Not yet implemented !');
			end
			end

			case 'Bootstrap'
			CM_stat = [];
			end
			end

Scripts_Sharing/ISFC/Utils/SlidingWindow.mbdfe7aadccccmaster

SlidingWindow.mView Options

Scripts_Sharing/ISFC/Utils/SlidingWindow.m
bdfe7aadccccmaster

SlidingWindow.m
View Options