See also nirs.modules.PCAFilter
nirs.modules.BaselinePCAFilter
This module runs PCA from baseline data.
PCA is a prefiltering method to reduce spatial covariance in the fNIRS data and works on the assumption that the systemicphysiological signals can be identified by a strong spatial covariance structure. As introduced by Zhang et al. [1], the first several principal components of the spatial covariance of the fNIRS data are removed to reduce global signals. In the oroginal proposed version of this, a separate baseline-only data file is used to derive the components, which are then projected from a separate data file of interest:
where U is the left singular vectors returned as the columns of a matrix; S is the singular values returned as a non-negative diagonal matrix in decreasing order; V is the right singular vectors returned as the columns of a matrix, those U, S, and V computed by singular value decomposition; Ybaseline and Ytask are the vectors of measurements from resting data as a baseline and task; and i is the index of number components.
The assumption in this model is that spatial structure due to the systemic noise is the same between the baseline-only and task file and therefore a filter can be designed from the baseline file and applied to the task file of interest. This is applied separately for the oxy- (HbO) and deoxy-hemoglobin (Hb) signals since these may have differing physiological contributions and therefore differing spatial structure. In Eq. (1b), n is the number of spatial components to be removed. Following the recommendation by Franceschini et al., [2], we suggest setting n to capture 80% of the spatial covariance by removing the first few eigenvectors of the baseline signal. Overal-removal could suppress brain signals and introduce type-II error.
Example Usage (1):
clear all
clc
for i = 1:4
raw(i,1) = nirs.testing.simData([], @(t)nirs.testing.blockedStimDesign(t,15,25,1));
% task-based 15s duration, 25s between onset, and 1 condition
end
raw(1).demographics(‘subject’) = ‘S1’; raw(1).demographics(‘task’) = ‘Tapping’;
raw(2).demographics(‘subject’) = ‘S2’; raw(2).demographics(‘task’) = ‘Tapping’;
raw(3).demographics(‘subject’) = ‘S1’; raw(3).demographics(‘task’) = ‘rest’;
raw(4).demographics(‘subject’) = ‘S2’; raw(4).demographics(‘task’) = ‘rest’;
nirs.createDemographicsTable(raw)
job = nirs.modules.KeepStims();
raw(3:4) = job.run(raw(3:4)); %Raw(3:4) are rest (do not have stim-mark)
%% Basic pipelines
job = nirs.modules.OpticalDensity;
job = nirs.modules.BeerLambertLaw(job);
job = nirs.modules.Resample(job);
hb = job.run(raw);
% Baseline PCA
job = nirs.modules.BaselinePCAFilter();
job.nSV=.8;
Task = find(ismember(nirs.createDemographicsTable(hb).task,‘Tapping’));
Baseline = find(ismember(nirs.createDemographicsTable(hb).task,‘rest’));
job.table = table(Task,Baseline);
hb_baselinePCA = job.run(hb);
Example Usage (2):
This example show how to use baseline PCA when the baseline/resting data is not available. Baseline/resting will be used from the condition rest from the file.
% Baseline PCA
job = nirs.modules.BaselinePCAFilter();
job.nSV=.8;
hb_baselinePCA2 = job.run(hb(find(ismember(nirs.createDemographicsTable(hb).task,‘Tapping’))));
References
[1] Zhang Y, Brooks DH, Franceschini MA, Boas DA. Eigenvector-based spatial filtering for reduction of physiological interference in diffuse optical imaging. J Biomed Opt. 2005 Jan-Feb;10(1):11014. doi: 10.1117/1.1852552. PMID: 15847580. https://pubmed.ncbi.nlm.nih.gov/15847580/
[2] Franceschini MA, Joseph DK, Huppert TJ, Diamond SG, Boas DA. Diffuse optical imaging of the whole head. J Biomed Opt. 2006 Sep-Oct;11(5):054007. doi: 10.1117/1.2363365. PMID: 17092156; PMCID: PMC2637816. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2637816/