Pipeline EEG

  • Acquisition
  • Pre-processing
  • Spindle detection
  • Spindle analysis
  • Reconstruction

Acquisition (CoRe)

  • Saturation of extra ECG fixed
  • Checking and saving impedance measures before and after each task- and sleep-EEG recordings -> quality of EEG signal for later analysis
  • Size and reference of used EEG caps saved -> building cap templates for each size using database from Paris

Preprocessing:

MRI + registration

  • Freesurfer
  • Non-linear registration T1 to MNI152
  • Inverse warp a cap template to subject space

 

 

EEG pre-processing

  • Gradient artifact removal (FASTR -  Niazy et al., 2005, NeuroImage)
  • ECG QRS-peak detection: automatic channel selection
  • Downsampling (250Hz)
  • BCG correction (FASST; ICA-based algorithm)
  • Bandpass filter (0.5-25Hz)
  • Re-referencing to averaged mastoids (M1-M2)
  • Extra Fast Kernel ICA (removal of movement-related artifacts)
  • Bad interval detection
  • ​Automatic script (Matlab):
    • Wavelet-based algorithm (Warby et al. 2014 Nature Methods) ​Github: https://github.com/Mensen/swa-matlab
    • Detect Spindles
    • Remove spindles detected during Bad Intervals
    • Add sleep scoring to spindle detection output
    • Write VMRK (marker file - brainvision)
      • Extract specific spindles depending on sleep stage
      • Add sleep scoring
    • Github: https://github.com/arnaudbore/spindlesDetection

Spindle detection

Matlab structure

  •     Ref_Region: electrodes
  •     Ref_TypeName: Fast or Slow spindle
  •     Ref_Start: Beginning of the spindle
  •     Ref_End: End of the spindle
  •     Ref_NegativePeak
  •     Ref_PositivePeak
  •     Ref_Peak2Peak: spindle amplitude
  •     Ref_Length: spindle duration
  •     Ref_NumberOfWaves: determination of spindle frequency
  •     scoring: Sleep Stage

Spindle detection - Outputs

Spindle detection - Outputs

Manual validation

Spindle analysis

  • Spindle features
    • Type (slow vs. fast, NREM2 vs. NREM3)
    • Number
    • Amplitude
    • Duration
    • Frequency
    • Density
  • Topographic distribution and propagation
  • Connectivity (coherence & phase-locking )
  • Hemispheric differences in spindle features and power (e.g., C3-C4; Nishida & Walker, 2007, PloSOne)

Reconstruction

  • Brainstorm  (Tadel et al., 2011, Comp. Int. & Neuro.)
  • BEM head model
  • Import EEG epochs based on spindle markers (-600 to 2000ms)
  • Remove DC offset (-600 to -100ms)
  • Noise covariance estimation
  • Bandpass filtering (spindle range: 11-17Hz)
  • Source estimation (minimum norm)
  • Signal projection and power estimate
  • Dynamic visualization (movie)
  • Deep sources?

Single spindle-event detected on Pz

Time-frequency maps (detected on Pz)

Pz

Fz

C4

% increase in the specific frequency band

Time-frequency maps (detected on Pz)

% increase in the delta frequency band

% increase in the spindle frequency band

Connectivity - Coherence (detected on Pz)

Pz - spindle band

Fz - spindle band

Fz - theta band

Pz - theta band

Connectivity - Coherence (detected on Pz)

Pz - spindle band

Fz - spindle band

Fz - theta band

Pz - theta band

T7 - spindle band

All channels

Pipeline EEG

By Arnaud Boré

Pipeline EEG

Presentation Pipeline

  • 592