TMFC_denoise is a MATLAB toolbox for SPM12/SPM25 that performs GLM-based denoising (noise regression).
This toolbox allows you to add noise regressors to the original general linear model (GLM), calculate framewise displacement (FD), Derivative of root mean square VARiance over voxelS (DVARS), and FD-DVARS correlation before and after denoising.
The updated GLMs can be used for task-activation analysis or task-modulated functional connectivity (TMFC) analysis.
For detailed documentation, please see: https://tmfc-denoise.readthedocs.io
For examples of how to use TMFC_denoise via the command line, please see the examples folder.
TMFC_denoise is available as a separate toolbox and as part of the TMFC toolbox (https://github.com/IHB-IBR-department/TMFC_toolbox).
- Add SPM12/SPM25 to your MATLAB path.
- Add TMFC_denoise OR TMFC_toolbox to your MATLAB path (Home --> Set Path --> Add with Subfolders --> Select TMFC_denoise OR TMFC_toolbox folder).
- Enter TMFC_denoise in the command window to open the TMFC_denoise GUI
or
Enter TMFC in the command window to open the TMFC toolbox GUI, then press Tools --> Denoise button.
TMFC_denoise can be run via GUI or the command line. To run TMFC_denoise via the command line, see TMFC_denoise.m function.
-
(1) Calculates head motion parameters (temporal derivatives and quadratic terms) (6HMP, 12HMP, 24HMP). Temporal derivatives are computed as backward differences (Van Dijk et al., 2012). Quadratic terms represent 6 squared motion parameters and 6 squared temporal derivatives (Satterthwaite et al., 2012).
-
(2) Calculates framewise displacement (FD) as the sum of the absolute values of the derivatives of translational and rotational motion parameters (Power et al., 2012).
-
(3) Creates spike regressors based on a user-defined FD threshold (Spike Regression). For each flagged time point, a unit impulse function is included in the general linear model; it has the value 1 at that time point and 0 elsewhere (Lemieux et al., 2007; Satterthwaite et al., 2012).
-
(4) Creates eroded WM and CSF masks.
-
(5) Creates aCompCor regressors (Behzadi et al., 2007). Calculates either a fixed number of principal components (PCs) or a variable number of PCs that explain 50% of the signal variability separately for the eroded WM and CSF masks (Muschelli et al., 2014).
-
(6) Creates WM/CSF regressors (Fox et al., 2005). Calculates the average BOLD signal separately for eroded WM and CSF masks. Optionally calculates derivatives and quadratic terms (Parkes et al., 2017) (2PHYS, 4PHYS, 8PHYS).
-
(7) Creates global signal regressor (Fox et al., 2005, 2009). Calculates the average BOLD signal for the whole-brain mask. Optionally calculates derivatives and quadratic terms (Parkes et al., 2017) (GSR, 2GSR, 4GSR).
-
(8) Calculates DVARS (Derivative of root mean square VARiance over voxelS), defined as the RMS of the differentiated BOLD time series within the GM mask (Muschelli et al., 2014). Also computes FD-DVARS correlations. DVARS is computed both before and after noise regression (for the original and updated GLM, respectively).
-
(9) Adds noise regressors to the original model and estimates the updated model. The noise regressors and the updated model are stored in the TMFC_denoise subfolder.
-
(10) Optionally applies robust weighted least squares (rWLS) for model estimation (Diedrichsen & Shadmehr, 2005). It assumes that each image has its own variance parameter; some scans may be disrupted by noise (high variance). In the first pass, SPM estimates the noise variances; in the second pass, each image is reweighted by the inverse of its variance.