Scribd Logo
Upload

Welcome to Scribd!

  • 77 views

    Lecture 2: EEG Basics: Introduction To Modern Brain-Computer Interface Design Christian A. Kothe SCCN, Ucsd

    Uploaded by

    Gilberto Perpiñan
    This document provides an overview of EEG basics for brain-computer interface design, including the underlying neural processes measured by EEG, the spatial characteristics of EEG signals, temporal characteristics such as event-related potentials and oscillatory processes, more complex EEG phenomena, sources of non-brain artifacts, and considerations for EEG sensor design and signal acquisition.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Lecture 2: EEG Basics: Introduction To Modern Brain-Computer Interface Design Christian A. Kothe SCCN, Ucsd

    Uploaded by

    Gilberto Perpiñan
    77 views54 pages
    This document provides an overview of EEG basics for brain-computer interface design, including the underlying neural processes measured by EEG, the spatial characteristics of EEG signals, temporal characteristics such as event-related potentials and oscillatory processes, more complex EEG phenomena, sources of non-brain artifacts, and considerations for EEG sensor design and signal acquisition.

    Original Title

    02_EEG_Basics.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document provides an overview of EEG basics for brain-computer interface design, including the underlying neural processes measured by EEG, the spatial characteristics of EEG signals, temporal characteristics such as event-related potentials and oscillatory processes, more complex EEG phenomena, sources of non-brain artifacts, and considerations for EEG sensor design and signal acquisition.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    77 views54 pages

    Lecture 2: EEG Basics: Introduction To Modern Brain-Computer Interface Design Christian A. Kothe SCCN, Ucsd

    Uploaded by

    Gilberto Perpiñan
    This document provides an overview of EEG basics for brain-computer interface design, including the underlying neural processes measured by EEG, the spatial characteristics of EEG signals, temporal characteristics such as event-related potentials and oscillatory processes, more complex EEG phenomena, sources of non-brain artifacts, and considerations for EEG sensor design and signal acquisition.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 54
    At a glance
    Powered by AI
    Some of the key takeaways are that EEG detects large-scale neural dynamics like event-related potentials and oscillatory processes, and the largest contributors are pyramidal cells in the cortex. It also discusses underlying brain processes, spatial and temporal characteristics.

    The two major BCI-detectable EEG phenomena are event-related potentials and oscillatory processes.

    Some of the major non-brain artifacts that can appear in EEG signals are muscle artifacts from neck/face/eye muscles, eye blinks, line noise, and sensor-related artifacts like drifts and noise.

    Lecture 2: EEG Basics

    Introduction to Modern Brain-Computer Interface


    Design

    Christian A. Kothe
    SCCN, UCSD
    Outline
    1. Underlying Brain Processes
    2. Spatial Characteristics
    3. Temporal Characteristics
    4. Complex EEG Phenomena
    5. Non-Brain Artifacts
    6. Sensing and Acquisition
    2.1 Underlying Brain Processes
    Underlying Neural Processes
    • All BCIs have to operate on observable effects
    of brain activity
    • Except for fMRI and fNIRS, they operate on
    effects of neural firing processes
    • EEG, MEG and ECoG can only
    detect large-scale neural
    dynamics
    • For example, 50.000 neurons
    firing in near-synchrony
    Underlying Neural Processes
    • Largest contributors to the EEG are the
    pyramidal cells
    • Radially oriented in the cortex (orthogonal to
    the surface)
    Pyramidal cell

    Hand et al., 2005


    Underlying Neural Processes
    • Largest contributors to the EEG are the
    pyramidal cells
    • Radially oriented in the cortex (orthogonal to
    the surface)
    Pyramidal cell
    • Electromagnetic fields of +
    ++
    co-aligned and co-activated ++
    - -- +
    neurons add up --
    --
    --

    Hand et al., 2005


    Large-Scale Brain Processes
    • When would 50.000 neurons fire near-
    synchronously?
    – An external event triggers a cascade of related
    neural processes (e.g., in perception)
    – An internal event triggers a cascade of related
    neural processes (e.g., sudden “aha!”)
    – Neural populations enter a synchronized steady-
    state firing pattern (e.g., idle oscillations)
    • Event-Related Potentials (ERPs) and Oscillatory
    Processes are the two major BCI-detectable
    EEG/MEG phenomena
    Signal Detectability
    • Root cause might not be directly observable (e.g.,
    dopaminergic system, deep brain structures, few
    neurons)
    • Widely scattered neural
    populations are unlikely to
    exhibit synchrony (unless
    connected by fiber tracts)
    • Spatially compact
    populations are more likely
    to have coordinated timing
    • Electromanetic fields can
    cancel each other out
    (e.g., in the Amygdala)
    Makeig 2007
    2.2 Spatial Characteristics
    Anatomical Regions
    • Some notable large-scale brain features are
    the hemispheres, lobes, gyri and suci

    Krepki, 2004
    Functional Mapping
    • For most regions more or less well known
    functional associations exist – the motor
    cortex is one of the best examples:

    Krepki, 2004
    Volume Conduction
    • Neural activity is conducted through the brain
    volume to the scalp and sensors by Volume
    Conduction
    • Volume conduction
    is linear
    • Each sensor measures a
    (weighted) sum of
    each neuron’s activity
    Volume Conduction
    • Note: the point-spread function from a source
    patch to the scalp is extremely broad

    Source Patch

    Akalin Acar et al., 2011


    Measurement Sites
    • Standardized location system (10-20 system)
    • Saves a lot of hassle vs. custom labels

    Malmivuo and Plonsey, 1995


    Measurement Sites
    • Defined also at much higher resolutions (not
    shown here)

    Malmivuo and Plonsey, 1995


    Actual Scalp Maps
    • Scalp maps (observed voltages at scalp sites) allow for source
    localization
    • Single-source scalp maps are rarely observed in the raw signal,
    but can be obtained by signal processing (more later)

    3d head visualization Corresponding 2d scalp maps


    Equivalent Dipole Model
    • Electromagnetic field sustained by a compact
    collection of neurons (e.g., 1 cm2) can be
    modeled as a single equivalent dipole
    • This facilitates localization of the field source
    Equivalent Dipole Tour
    • Further scalp maps and associated dipole fits
    Dipole Modeling Problems
    • High-quality fits are hard to achieve
    – Requires knowledge about sensor locations
    – Requires assumptions about conductivities of scalp,
    skull, cerebrospinal fluid (CSF), brain tissue
    – Requires knowledge of the folding of the cortex
    (candidate dipoles) unless simplistic spherical model is
    used
    – Some brain tissue has anisotropic conductance (white
    matter)
    – Scalp maps are usually not perfect (arise from data
    processing) – fit accuracy suffers
    – Scalp maps can be a sum of multiple dipole sources –
    requires a distributed source model
    Distributed Source Modeling
    • Allow to recover and image distributed cortical
    support of given scalp maps

    Subsequent images:
    Rey Ramirez (Scholarpedia)
    Distributed Source Modeling
    • Wide range of methodologies and underlying
    assumptions (sLORETA, Beamforming, Sparse
    Bayesian Learning, …)
    • Prone to finding only locally optimal solutions

    LCMV Beamforming Anatomically Constrained Beamforming


    Distributed Source Modeling
    • Wide range of methodologies and underlying
    assumptions (sLORETA, Beamforming, Sparse
    Bayesian Learning, …)
    • Prone to finding only locally optimal solutions

    sLORETA Sparse Bayesian Learning


    Distributed Source Modeling

    (Video)
    2.3 Temporal Characteristics
    Neural vs. Scalp Activity
    • Typical spiking behavior of a single neuron
    25ms

    Tsodkys, 1997

    • Typical signal measured at a scalp site

    250ms
    Event-Related Potentials (ERPs)
    • Averaging EEG activity relative to an event
    results in primarily event-induced activity
    (trial-to-trial variability averaged out)
    Event-Related Potentials (ERPs)
    • Single-trial ERPs are much harder to identify

    Stimulus presented Person commits an error


    Oscillatory Processes
    • EEG is permeated by oscillatory processes,
    such as the alpha rhythm (pictured)
    • Standard names for such rhythms are delta (0-
    4Hz), theta (4-7Hz), alpha (8-13Hz), beta (12-
    30Hz), and gamma (25-100Hz)

    Kaplan et al., 2000


    Oscillatory Processes
    • Alpha: Sensory areas (visual cortex, auditory
    cortex) and Motor areas (motor cortex) exhibit
    strong alpha-band oscillations when “idle” in
    most subjects
    • Beta: Motor cortex often generates also beta-
    band oscillations
    • Theta: Known to occur in “bursts” relative to
    events in certain brain areas (e.g. frontal
    midline, lateral frontal, …)
    2.4 Complex EEG Phenomena
    EEG Sources Separated via ICA
    Higher-Order Phenomena
    • Event-Related Coherence between two signal
    components (here simulated data)

    Makeig 2007
    Effective Connectivity
    • Sophisticated measure of interaction between
    multiple signals (“information flow”)

    Mullen et al., 2011


    Effective Connectivity

    (Video)
    2.5 Non-Brain Artifacts
    Non-Brain Artifacts
    • Often far outscale the brain processes in the
    EEG (when present)
    • Internally generated: neck, face and eye
    muscles, eye dipoles, heart activity
    • Externally generated: 50/60Hz line noise, EM
    spikes from equipment
    • Sensor-related: DC offset drifts, cable sway,
    thermal noise, quantization noise
    Muscle Artifacts
    • High-frequency / broadband, large amplitude

    1s
    Muscle Artifacts
    • Scalp projections are spatially stereotyped
    Eye Blinks
    • Large low-frequency peak and rebound, mainly
    frontal
    • Can also incurs non-linear effects in occipital cortex

    1s
    2.6 Sensing and Acquisition
    EEG Sensor Designs
    • Most EEG systems are gel-based
    • Nowadays mostly using active electrodes

    Passive, gel-based Active, gel-based


    (EasyCap) (Brain Products)
    EEG Sensor Designs
    • Dry (gel-free) systems are emerging quickly

    Pins Spring-loaded Pins Foam-based sensors


    (g.SAHARA) (NCTU) (NCTU)
    EEG Sensor Designs
    • Recent Prototypes

    Bristle Sensors Epidermal Electronics


    (Grozea et al., 2011) (Kim et al., 2011)
    Digitization
    • After amplification (e.g., 50000x), signal is
    low-pass filtered using an analog filter, then
    digitally sampled at fixed rate

    Ottewill
    Sampling Theorem
    • If the signal is band-limited below the Nyquist
    frequency B (i.e., contains no higher frequency
    than B), it can be exactly reconstructed using
    the interpolation function:

    • The Nyquist Frequency is ½ sampling rate


    Computer-based Access
    • The data is made accessible through:
    – Vendor-specific recording programs (BrainVision
    Recorder, ActiView, g.Recorder, …)
    – Vendor-specific system drivers (Emotiv SDK,
    BioSemi driver, …)
    – Generic system interfaces (e.g., Bluetooth serial
    port, A/D cards, TCP, …)
    • Almost all EEG systems support real-time
    signal access (except for some gadgets)
    L2 Questions?

    You might also like