EEG and Eye-Tracking Preprocessing Steps

  • preprocessed by MATLAB
  • triggers and latencies extracted and specific temporal order of patients determined 
  • replace and identify bad electrodes (electrodes that are > 3 stdev away from other electrodes in frequency spectrum distribution) using spherical spline interpolation (?)
  • noisy channels filtered - either visually or replaced entirely by zeros
  • 109 channels for EEG,  6 EOG for artifact removal

More Preprocessing

  • data high pass filtered at 0.1 Hz and notch filtered at 59-61 Hz
  • Principal Components Analysis (PCA) algorithm removed sparse noise - recovers a low-rank matrix from a corrupted matrix by subtracting the error.
  • visual inspection after preprocessing as well.

Tools for EEG Analysis

  • EYE-EEG: plugin for MATLAB- integrates analysis for electrophysiological + eye-tracking data.
  • adds eye-tracking data as extra channels to EEG
  • Adaptive velocity-based algorithm to detect saccades and fixations, i.e. a blink
    • identifies fixation as groups of consecutive points
    • uses that to map minimum and maximum x and y values - compare that to maximum dispersion value.
    • if the dispersion is lower, then its a fixation. 

Misc about Data 

  • Code Availability: analysis performed in MATLAB + EEGlab
  • Data records: a folder for each subject. an EEG binary file, two eye-tracker files. One is segmented into blinks, saccades, and fixations. The other isn't. Each folder has a file for each paradigm.
  • Behavioral Assessment: self-report by Collaborative Informatics and Neuroimaging Suite (COINS)
  • Cognitive Assessment: two research assistants obtained separate raw scores for each patient, entered as dataset
  • Developmental History: medical/school/demographics

Resting EEG Analysis

  • after filter, data segmented into eye-closed and eye-open segments
  • analyze eye-closed
  • EEG recomputed against average reference, split into 2-second epochs, then Fourier transformed
  • Resting-state EEG had their spectral amplitude saved. 
  • Spectra for each channel averaged for the epochs
  • group mean spectral amplitude for all or each electrode calculated
  • split up into the different frequency bands (alpha beta etc)

Sequence Learning Paradigm

  • EEG data segmented based on stimulus of each filled white circles at each of the eight different locations
  • 900 ms long epochs (100 ms pre, 800 ms post)
  • averaged trial types (hits, missed, learned) for each subject individually and calculated group average
  • found a decrease of P300 over different blocks
  • behavioral performance: percentage of correctly learned spatial location
  • learning rate: newly learned location in relation to all possible locations

More Sequence Learning Paradigm

  • P300 amplitude = amplitude on electrode CPZ, with latency of 350-500 ms post stimulus
  • slightly delayed because children have greater latency
  • performance increases while learning rate and P300 decrease w practice
  • recall the expectation that subjects learn to expect the order of the appearance of targets w/ 5 blocks
  • decrease in P300 because of novelty indicator of a target stimulus- subjects are not surprised anymore!

WISC Symbol Search / Processing Speed Paradigm

  • Eye-tracking data
  • recall the goal is to find two target symbols among a set of five search symbols
  • graphic - three sub regions: target, search group, response buttons
  • calculated number of saccade steps, repetitions, pupil size, and protracted gaze dwell time (fixation duration) for each sub region
  • graph to indicate the duration of fixation and distribution

Naturalistic Stimuli Paradigm

  • covariance matrices were computed 
  • found the 3 strongest correlated components and compute corresponding correlation values
  • "how does this specific video probe people's EEG responses"
  • ISC measure

Surround Suppression Paradigm

  • extracted based on flickering stimulus
  • 4 foreground contrast (0%, 30%, 60%, 100%) and 3 background conditions (parallel, orthogonal, none)
  • data of two blocks merged for each subject
  • computed SSVEP signal without background, and a SSVEP averaged all conditions with a background
  • FFT computed to obtain a measure of SSVEP power at 25 Hz freq
  • subtracted this from neighboring freq to get the actual evoked activity

More on Surround Suppression Paradigm

  • in-house algorithm to detect highest SSVEP amplitude, compute signal based on the average of max electrode + four surrounding electrodes
  • graph to demonstrate the difference in SSVEP amplitude between different contrasts with or without a background
  • increase SSVEP amplitude with an increase in foreground contrast
  • relative reduction in SSVEP amplitude due to surround contrast

Contrast Change Detection Paradigm

  • data of three blocks merged for each subject
  • target epochs extracted from 500 ms before target to 1000 ms after peak sensory evidence
  • response-locked epochs extracted from 1000 ms before repsonse to 300 ms after response (buttons)
  • rejected trials for scalp channel over 100 microV
  • SSVEP based on stimulus locked epochs + motor response signal based on response-locked epochs measured by Fourier
  • CPP analysis averages single-trial  waveforms (baseline corrected 500 ms before response)

More on Contrast Change Detection Paradigm

  • Figure to show SSVEP, CPP, and motor response signal topographies
  • Result shows posterior maximum for SSVEP around electrode Oz. 
  • CPP shows highest activity near CPz electrode and motor response signal peaks over C3 and C4.
Made with Slides.com