ASCT manual
  • About ASCT
  • Getting Started
    • Installation
    • Introduction
  • Manual
    • Run Control
    • Data import
    • Preprocessing
    • Experimental design
    • Sensor space AR (AR1)
    • ICA decomposition
    • ICA-based AR (AR2)
    • ICA2 and classification
    • Source localization
  • Signal Reconstruction
  • Connectivity estimation
  • Statistics, visualization, and data export
  • Appendix
Powered by GitBook
On this page
  1. Manual

Preprocessing

PreviousData importNextExperimental design

Last updated 2 months ago

Filtering

The windowed-sinc FIR filters are recommended for analyzing EEG/MEG signals . As the connectivity estimation is typically performed in a frequency range between theta and beta, high-pass (HP) filtering is recommended to be set just below this range (1-2 Hz typically). Removing very low-frequency spectral components that dominate the EEG/MEG spectral power will help in further ICA decomposition and MVAR model fitting. At the same time, MVAR modelling requires some amount of noise in the signal, and the low-pass (LP) filter cut-off should be set just below the line frequency (50 or 60 Hz, depending on location). Filter order can be estimated with the following formula that links the filter transition band and sampling frequency:

Pf = 3.3 * fs / Wt (Pf - filter order, fs - sampling frequency, Wt - the width of the filter transition band).

Recommended filter settings are:

  • high-pass 2 Hz cutoff with a transition bandwidth of 0.5-1 Hz;

  • low-pass 48 Hz cutoff with a 2-3 Hz transition bandwidth.

All these parameters can be set in the filter section. Using windowed-sinc FIR filters is obtained by setting relevant HP and LP parameters to 'firws'. 

filter.HPfreq = 2;        % high pass frequency
filter.HPtype = 'firws';  % hp filter type
filter.HPord  = 6766;     % hp filter order
filter.LPfreq = 48;       % lp frequency
filter.LPtype = 'firws';  % lp filter type
filter.LPord  = 1000;     % lp filter order   
filter.plotResp   = 'no'; % show filer response
filter.resampleTo = 128;  % specify new sampling rate at preprocessing

Downsampling

Additionally, after filtering, the signal can be downsampled with the filter.resampleTo parameter. Recommended resulting rate around 250-256 Hz is recommended.

Optional downsampling is performed after filetring, so the filter order calculations need to take into account the raw file sampling rate.

Temporal oculomotor correction (EEG only)

For optimal performance of further artifact rejection, an additional step of ocular correction is performed. This is only possible when the EOG signal is available for EEG recording. EOG correction is performed with the RLS method (). EOG-corrected version of the signal is used for further detection of artifact, which benefits from removing EOG bias on frontal electrodes.

Saving preprocessed data

After preprocessing, datasets are saved in the 1_PREPROCESSED directory with the _PREP.mat extension. FieldTrip format is used here, supplemented with some additional fields.

If EOG correction is available, EOG-corrected data are also saved along with the uncorrected signal _PREP_EOGcorr.mat extension.

Matlab uses compression of mat files by default which saves a space but significantly increases write time. To disable compression, rc.saveCompress option needs be set to zero.

(Widmann, 2012)