Skip to content

Event-Notes-Day3.md

Latest commit

 

History

History
138 lines (92 loc) · 6.22 KB

Event-Notes-Day3.md

File metadata and controls

138 lines (92 loc) · 6.22 KB
Raw

📝 Event Notes – Day 3

Day 3 – Tuesday, 10 September 2024

EEG Signal Processing Techniques and Feature Extraction

Session 3: EEG Signal Processing Techniques and Feature Extraction — Dr. Abdelkader Nasreddine

Key Takeaways:

  • Computational Intelligence for EEG-based BCI:

    • Pre-processed EEG signals → feature extraction → classifier → BCI output.
    • Feature extraction is critical: poor features → inaccurate classification.

  • Feature Types:

    • Temporal (Time) Features: Based on signal amplitude over time, e.g., P300.
    • Spectral (Frequency) Features: Band-power analysis, e.g., SSVEP; isolate via band-pass filter.
    • Time-Frequency Features: Combines time and frequency; e.g., wavelet analysis for motor imagery (MI).

  • Feature Classification in BCI:

    • Converts extracted features into device control commands (continuous or discrete).
    • Linear Models: Linear Discriminant Analysis (LDA) – efficient, low computational cost, suitable for online BCI.
    • Non-linear Models: Artificial Neural Networks (ANNs) – can capture complex, non-linear relationships; learns patterns without predefined features.

  • Sensor-level vs Source-level EEG Analysis:

    • Sensor-level (Scalp electrodes): Direct, efficient, captures overall patterns; limited by volume conduction.
    • Source-level: Localizes brain activity sources; precise but computationally intensive and noise-sensitive.

  • Post-processing & Data Epoching:

    • Continuous EEG → segmented into epochs [electrodes × time × trials].
    • Time-locked vs Phase-locked: Signal changes at same time or phase across trials.
    • Evoked vs Induced Activity: Evoked → stimulus-driven, phase-locked; Induced → top-down processes, not necessarily phase-locked.

  • ERP (Event-Related Potentials):

    • Time-locked response to stimuli; average across trials to enhance SNR.
    • Characterized by amplitude, latency, polarity, and scalp distribution (e.g., P300).

  • ERO (Event-Related Oscillations):

    • Frequency-specific changes (theta, alpha, beta, gamma); often analyzed via time-frequency techniques.

  • Motor & Mental Imagery:

    • Motor imagery → changes in sensorimotor frequency bands.
    • Mental imagery → visualizations or abstract concepts → induced oscillatory changes.

Tutorial 3: EEG Temporal and Spectral Analysis — Dr. Mohamed Zaky

Key Takeaways:

  • ERP Analysis Workflow:

    • Epoch extraction → baseline correction → channel visualization → manual/automatic artifact rejection.
    • Tools: EEGLAB GUI (Plot > Channel data, Plot > Channel ERPs, ERP scalp maps).
    • Components: peak amplitude, mean amplitude, peak latency, topographic mapping.

  • Group-level Analysis:

    • Multi-level approach: first level → individual subjects; second level → groups.
    • Data organized using BIDS (Brain Imaging Data Structure).
    • Preprocessing: clean raw data, ASR, ICA, re-reference, channel interpolation.
    • Study design setup → precompute measures → visualize ERPs per channel or across all channels.

  • SSVEP Feature Analysis:

    • Frequency domain analysis (Fourier Transform) to extract power at stimulus frequencies.
    • Classification via LDA, SVM, or deep learning.
    • ICA can separate SSVEP from other brain activity.

  • SSVEP Datasets Used for Practice:

    • Dataset 1: 12-target visual stimuli, 8 occipital channels (BioSemi ActiveTwo), 256 Hz sampling.
    • Dataset 2: BETA database, 64 channels, 250 Hz sampling, 40 characters flickering 5.8–8 Hz.
    • Dataset 3: MAMEM SSVEP dataset, 14 wireless channels (Emotiv Epoc), 128 Hz sampling, 5 simultaneous flicker frequencies.

Workshop 3 – Hands-on EEG Data Recording & Analysis (SSVEP Task)

Speakers: Dr. Mohamed Zaky, Eng. Mai Gamal Focus: Practical EEG recording, preprocessing, ERP & SSVEP analysis

Day 1–2 Recap: EEG Recording & Preprocessing

  • EEG Setup: g.tec Unicorn Hybrid Black, 8 dry electrodes, wireless via Bluetooth

  • Electrode Placement: 10–20 system, impedance checked, proper scalp contact

  • Data Acquisition: Record SSVEP responses while focusing on visual stimuli

  • EEGLAB/MATLAB Preprocessing Steps:

    1. Load raw EEG + event markers
    2. Filtering: high-pass (~0.5 Hz), notch (50 Hz)
    3. Re-reference
    4. Artifact removal: ASR + ICA

Day 3 – Practical EEG Analysis

1. Epoch Extraction & Baseline Correction (Day three starting workshops)

  • Tools > Extract epochs → select events “square” & “RT”
  • Tools > Remove epoch baseline → detect relative changes vs baseline

2. Visualization – Subject Level

  • Plot > Channel data (scroll) → inspect each channel
  • Reject bad epochs → manually or using ERPLAB
  • Plot > Channel ERPs > With scalp maps → averaged ERPs + scalp distribution
  • Plot > Channel ERPs > In scalp/rect array → visualize single channel trace
  • Plot > ERP map series > In 2-D → scalp maps at specified latencies

3. Group-level ERP Analysis

  • Import BIDS dataset → File > BIDS tools > Import BIDS folder to STUDY
  • Edit data → remove non-EEG channels + re-reference
  • Denoise → Tools > Clean Rawdata + ASR → interpolate → re-reference
  • ICA → Tools > Decompose data → classify (ICLABEL) → remove artifacts
  • Epoching & baseline correction → Tools > Extract epochs → Remove epoch baseline
  • Study design → Study > Select/Edit study design(s) → New (Independent variable)
  • Precompute channel measures → Study > Precompute channel measures
  • Visualization → Study > Plot channel measures (per channel or all channels)

4. SSVEP Feature Extraction & Analysis

  • Feature Extraction: Fourier Transform → power or correlation at stimulus frequencies
  • Classification: Choose classifier (LDA / SVM) → discriminate stimulus frequencies
  • Model Training & Evaluation: Train classifier → evaluate on test set
  • Advanced Techniques: ICA → separate SSVEP components; Deep Learning → feature extraction & classification

Outcome: Hands-on experience in EEG recording, preprocessing, epoch extraction, baseline correction, ERP/SSVEP analysis, group-level statistical visualization, and feature extraction for BCI applications