Добавил:
Sekretar
kiopkiopkiop18@yandex.ru
t.me/Prokururor I Вовсе не секретарь, но почту проверяю
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Ординатура / Хирургия / Библиотека им академика М.И. Перельмана / Книга_6027_Библиотеки_им_академика_М_И_Перельмана.pdf
X
- •Preface
- •Acknowledgments
- •About This Book
- •Contents
- •List of Figures
- •List of Tables
- •Editor and Contributors
- •1.3 EEG and Other Neuroscience Methods
- •1.4 The Future of EEG
- •1.1 EEG Technology: Past to Present
- •1.2 What Do We Know About the EEG Signal?
- •1.5 Conclusions
- •References
- •2.1 Physiological Origins of the EEG
- •2.2 Signals of the EEG
- •2.3 Concluding Summary
- •References
- •3.1 Introduction
- •3.2 General Organization
- •3.3 Finding Your Way Around: Brain Atlases
- •3.3.2 Talairach Atlas and MNI Coordinates
- •3.3.4 Accessing and Using Atlases
- •3.4 Putting into All Together
- •3.5 Conclusion
- •References
- •4.1 Introduction
- •4.2 Overview of the Peripheral Nervous System
- •4.3 Basic Anatomical Unit of the PNS: Ganglia and Nerves
- •4.4 Anatomy of Somatic Nervous System
- •4.4.1 Receptors
- •4.4.1.1 Vision
- •4.4.1.2 Audition
- •4.4.1.3 Vestibular System and Balance
- •4.4.1.4 General Sensory Modalities
- •4.4.2 Somatic Sensory System
- •4.5 Anatomy of the Autonomic Nervous System
- •4.5.1 Sympathetic Nervous System
- •4.5.2 Parasympathetic Nervous System
- •4.6 Cranial Nerves
- •4.7 Function of the PNS and CNS as a Unit
- •4.8 Concluding Remarks
- •References
- •5.1 Introduction
- •5.2 Head Anatomy and Signal Propagation
- •5.3.1 The Eyes and Ocular Potentials
- •5.3.2 Facial Muscles and EMG
- •5.3.3 Sweat Glands and Skin Potentials
- •5.3.4 Blood Vessels and Heartbeat
- •5.4 Conclusion
- •References
- •6.1 Introduction
- •6.2.1 What Is a Brain State?
- •6.2.2 Brain States Measured with EEG
- •6.3 Examples of Brain States
- •6.3.1 Awake State Sleep State
- •6.3.2 Consciousness States: Presence Loss (Anesthesia)
- •6.4 Pathological Brain States
- •6.4.1 Traumatic Brain Injury
- •6.4.2 ADHD
- •6.5 Framework of Brain States
- •6.6 Concluding Summary
- •References
- •7.1 Introduction
- •7.3 Identifying Task Processes Within a Trial Segment
- •7.3.2 Cross-Trial Comparability and Flexible Time-Locking
- •7.4 What Does Activity Look Like on the Timeline?
- •7.5 Conclusion
- •References
- •8.1 Introduction
- •8.2 Setting a Research Question
- •8.3 Setting a Hypothesis
- •8.3.2 Testing the Hypothesis
- •8.4 Design of the Study
- •8.4.1 Contextualization of the Hypothesis
- •8.4.1.1 Experimental Paradigm
- •8.4.1.2 EEG Index
- •8.4.1.3 Group/Sample
- •8.4.2 Implementation of the Study
- •8.4.2.1 Paradigm/Task Implementation
- •8.4.2.2 Measurement Precision
- •8.4.2.3 Experimental Protocol
- •8.4.2.4 Pilot Testing
- •8.5 Concluding Summary
- •References
- •9.1 Introduction
- •9.2 Why Is Statistics Needed in EEG Research?
- •9.3 When Is Statistics Applied During EEG Data Analysis?
- •9.3.1 Raw EEG Data
- •9.3.2 Individual-Level (First-Level) Analysis
- •9.3.3 Group-Level (Second-Level) Analysis
- •9.3.3.1 Statistical Hypotheses
- •9.3.4 Application of Statistical Inference
- •9.3.5 Interpretation and Inference
- •9.4.1 Hypotheses (Upper Plane of Fig. 9.2)
- •9.4.2 Population and Sample Data (Bottom Plane of Fig. 9.2)
- •9.4.3 Sample Statistic (Middle Plane of Fig. 9.2)
- •9.5 Conclusion
- •References
- •10.1.1 Why Pilot Testing Matters
- •10.2 How to Prepare and Run the Pilot Testing
- •10.2.1.1 Signal Quality
- •10.2.1.2 Task Parameters
- •10.2.1.3 Instructions
- •10.2.1.4 Participant Experience
- •10.2.1.5 Equipment Setup
- •10.2.1.6 Procedures
- •10.2.1.7 Questionnaires
- •10.1 What Pilot Testing Is
- •10.3 Concluding Summary
- •References
- •11.1 Introduction
- •11.2 Lab Management
- •11.2.1 Admin and Organisation
- •11.2.2 Hardware and Software Maintenance
- •11.2.3 Lab Logbook
- •11.3 Keep Your Own Lab Notebook
- •11.4.1 Pre-measurement
- •11.4.2 Measurement
- •11.4.3 Post Measurement
- •11.5 Conclusion
- •References
- •12.1 Introduction
- •12.2 Components of the System
- •12.2.1 Detecting the Signal: EEG Electrode Technology
- •12.2.1.1 Passive Electrode Plus Gel
- •12.2.1.2 Active Electrodes Plus Gel
- •12.2.1.3 Passive Electrode and Saline Soaked Sponges
- •12.2.1.4 Dry Electrodes
- •12.2.1.5 Electrode Positions
- •12.2.2 Detecting the Signal: Sensors for Other Measures
- •12.2.2.1 Bipolar Peripheral Electrophysiology
- •12.2.2.2 Peripheral Physiological Sensors
- •12.2.2.3 GSR
- •12.2.2.4 Respiration
- •12.2.2.5 Photoplethysmography (PPG)
- •12.3 Conclusion
- •References
- •13.1 Purpose and Features
- •13.2 Before Starting Your Study
- •13.2.1 General Parameters
- •13.2.2 Special Applications
- •13.2.3 Real-Time Processing
- •13.3 During a Measurement Session
- •13.4 Troubleshooting
- •13.5 Conclusion
- •References
- •14.1 Introduction
- •14.2 Importance of Triggers
- •14.3 Advantages of Triggers
- •14.4 Disadvantages of Triggers
- •14.5 Alternatives to Triggers
- •14.6 Good Practice for Using Triggers
- •14.7.1 Setup
- •14.7.2 Analysis
- •14.7.3 Interpretation
- •14.8 Conclusion
- •References
- •15.1 Introduction
- •15.2 The Idea of Signal-to-Noise Ratio (SNR)
- •15.3 Sources of Artifact
- •15.4 Common Physiological Artifacts
- •15.4.1 Eye Artifacts
- •15.4.2 ECG Artifacts
- •15.4.4 Other Physiological Artifacts
- •15.5 Common Technical Artifacts
- •15.5.1 Technical Artifacts
- •15.5.2 Electrode Artifacts
- •15.5.3 Gel-Related Artifacts
- •15.5.4 Movement Artifacts
- •15.5.5 Body/Head Movements
- •15.5.6 Cable Movement Artifacts
- •15.6 Artifacts in Advanced Applications and Multi-modal Recordings
- •15.6.1 EEG and Functional MRI
- •15.6.2 EEG and Non-invasive Brain Stimulation
- •15.7 Optimizing the EEG Recording Quality
- •15.7.1 Focus on the Cap Preparation
- •15.7.2 Optimize the Recording Environment
- •15.7.3 During the Recording
- •15.7.4 Post Recordings
- •15.8 Conclusion
- •References
- •16.1 Introduction
- •16.2 Lab Infrastructure
- •16.2.1 Signal Quality
- •16.2.2 Control Over the Experimental Environment
- •16.2.4 Safety
- •16.3 Position of the Equipment and Accessories
- •16.4 Lab Procedures
- •16.5.1 Mobile Setups
- •16.5.2 Electrode Types
- •16.5.2.1 Passive Sponge-Based Electrodes
- •16.5.2.3 Dry Electrodes
- •16.5.3 Special Populations
- •16.5.3.1 Children
- •16.6 Concluding Summary
- •References
- •17.1 Introduction
- •17.2 Common Preprocessing Steps: Data Transformation
- •17.2.1 Inspecting Data
- •17.2.2 Changing the Sampling Frequency
- •17.2.3 Re-referencing
- •17.2.4 Interpolating Channels or Data Portions
- •17.2.5 Segmenting Data
- •17.3 Common Preprocessing Steps: Artifact Handling
- •17.3.1 Filtering
- •17.3.2 Attenuating Artifacts
- •17.3.2.1 Independent Component Analysis (ICA)
- •17.3.2.2 Regression Techniques
- •17.3.2.3 Template Subtraction Methods
- •17.3.3 Rejecting Artifacts
- •17.5 Tools for Processing and Analyzing EEG
- •17.6 Concluding Remarks
- •References
- •18.1 Introduction
- •18.2 Characterizing an Oscillatory Process
- •18.2.1 Fundamental Characteristics of an Oscillatory Process
- •18.2.2 From Time to Frequency and Back
- •18.3 Foundation for Spectral Analysis: The Dot Product
- •18.4 Fourier Analysis
- •18.4.1 From Vectors to Sinusoids: The Fourier Connection
- •18.4.2 The Fourier Family
- •18.4.3 Discrete Fourier Transform
- •18.4.4 Power Spectrum
- •Further Readings
- •References
- •19.1 Introduction
- •19.2 How to Get from EEG to ERPs
- •19.2.1 How to Process Your ERP Data
- •19.2.1.1 Pre-processing
- •19.2.1.2 Trial Selection
- •19.2.1.3 Baseline Correction
- •19.2.1.4 Averaging
- •19.2.2 Interpreting ERPs
- •19.2.3 Group Analysis
- •19.2.4 Single-Trial Analysis
- •19.3 Characteristics of the ERP and Its Components
- •19.4 Commonly Investigated ERP Components
- •19.4.1 Early Sensory Components
- •19.4.2 Long-Latency Sensory Components
- •19.4.3 Later Cognitive Components
- •19.4.4 ERP Components in Multimodal Recording Scenarios
- •19.5 Extraction of ERP Features
- •19.6 Conclusion
- •References
- •20.1 Introduction
- •20.2 Fundamentals of EEG Source Imaging
- •20.3 Forward Problem
- •20.4 Source Estimation
- •20.5 Statistical Inference in the Source Space
- •20.6 Source Connectivity
- •20.7 Conclusion
- •References
- •21.1 Introduction
- •21.2 Raw Data Access
- •21.2.1 How to Get Raw Data
- •21.2.2 How to Work with Raw Data Online
- •21.2.3 What Factors to Consider for Online Processing
- •21.3 Designing an Online Processing Experiment, an Example
- •21.4 Conclusion
- •References
- •22.1 Introduction
- •22.1.2 Chapter Overview
- •22.2.2 Exactly What the SME Means
- •22.2.5 Why the Scoring Method Matters
- •22.2.8 Other Potential Uses of the SME
- •22.4 Metrics of Reliability
- •22.5 Final Thoughts
- •References
- •23.1 Cognitive Neuroscience
- •23.1.1 Neuropsychology and EEG
- •23.1.2 Mental Chronometry and EEG
- •23.2 Research on the EEG Signals
- •23.3 Conclusions
- •References
- •24.1 Introduction
- •24.2.1 Clinical Research
- •24.2.2 EEG in Research for Clinical Applications
- •24.2.3 EEG as a Biomarker
- •24.3 Examples of Clinical Applications of EEG
- •24.3.1 Epilepsy
- •24.3.2 Sleep and Sleep Disorders
- •24.3.3 Anesthesia
- •24.4.2 Brain-Computer Interfaces and Movement Disorders
- •24.5 Future of EEG in Clinical Applications
- •24.6 Conclusion
- •References
- •25.1 Introduction
- •25.1.1 Why Connect Brains and Computers?
- •25.1.2 What Is a BCI?
- •25.1.3 Types of BCIs: Active, Reactive, Passive
- •25.1.4 BCIs in Neuroscience and HCI
- •25.2 Signals and Sensors
- •25.2.1 Neural Signals for BCIs
- •25.2.2 Wearable EEG and Form Factors
- •25.3 The BCI Pipeline: From Raw Signals to Decisions
- •25.3.1 Overview
- •25.3.2 Experimental Design and Labeling
- •25.3.3 Preprocessing and Artifacts
- •25.3.4 Feature Extraction
- •25.4 BCI Types Illustrated
- •25.4.1 Motor Imagery as an Active BCI Paradigm
- •25.4.2 P300 and SSVEPs as Reactive BCI Paradigms
- •25.4.3 Workload, Error, and Other Passive BCI Paradigms
- •25.5.1 Mental State Assessment as a First Stage
- •25.5.2 Open- and Closed-Loop Adaptation
- •25.6 Practical Challenges
- •25.6.1 Mobility, Artifacts, and Non-stationarity
- •25.6.2 Cross-User and Cross-Session Generalization
- •25.6.3 Evaluation in Real Settings
- •25.6.4 Ethics, Privacy, and Neurorights
- •25.7 Conclusions and Outlook
- •25.7.1 Key Takeaways
- •25.7.2 Future Trajectories
- •References
- •26.1 Focal Epilepsy
- •26.2 EEG Manifestations of Focal Epilepsy
- •26.2.1 Ictal EEG Patterns
- •26.2.2 Interictal EEG Patterns
- •26.3 Localization of Ictal and Interictal EEG Events
- •26.4 Intracranial EEG in Presurgical Planning
- •26.5 AI in EEG Interpretation
- •26.6 Conclusion
- •References
- •27.1 Introduction
- •27.2 Neonatal EEG Applications
- •27.2.2 Somatosensory States Monitoring in Neonates
- •27.3 Paediatric EEG Applications
- •27.3.1 Sleep Monitoring in Children and Adolescents
- •27.4 Future Directions and Conclusion
- •References
- •28.1 What Is Sleep?
- •28.1.1 Stages of Sleep
- •28.1.2 How Sleep Changes with Age
- •28.2 Measuring Human Sleep
- •28.2.1 The Various Forms of Sleep
- •28.2.2 Unihemispheric Sleep
- •28.3.1 NREM Sleep and Learning
- •28.3.2 REM Sleep and Learning
- •28.4.1 Active Brain Networks During Sleep
- •28.4.2 Measuring the Balance of Excitation and Inhibition in the Human Brain
- •28.4.3 The Cerebrospinal Fluid Dynamics in Human Sleep
- •28.5 Conclusions
- •References
- •29.1 What Is Mobile EEG?
- •29.2 Range of mEEG Systems
- •29.3 Technical Considerations
- •29.4 Validation
- •29.5 Application
- •29.6 Mild Cognitive Impairment
- •29.7 mEEG and Health and Exercise
- •29.8 mEEG in Sports
- •29.9 Conclusions
- •References
- •30.1 Introduction
- •30.2 General Framework and System Overview
- •30.3 Spectrum of Studies
- •30.4 MoBI+ Framework
- •30.5 Processing Multimodal MoBI Data
- •30.6 Challenges and Limitations
- •30.7 Conclusion
- •Appendix
- •List: Traveling with MoBI Equipment
- •References
- •31.1 Introduction
- •31.2 Types of Electric Brain Stimulation
- •31.3 Online Effects
- •31.3.1 Conventional Artifact Removal Strategies
- •31.3.1.1 Transcranial Direct Current Stimulation (tDCS)
- •31.3.1.2 Transcranial Random Noise Stimulation (tRNS)
- •31.3.1.3 Transcranial Alternating Current Stimulation (tACS)
- •31.3.3 Innovative Approaches to Minimize Artifacts
- •31.3.3.1 Non-Sinusoidal Waveforms
- •31.3.3.2 Amplitude-Modulated tACS (AM-tACS)
- •31.3.3.3 Transcranial Temporal Interference Stimulation (tTIS)
- •31.3.3.4 Summary of Advantages and Limitations
- •31.4.1 Spectral Power
- •31.4.2 Phase Locking/Phase Coherence
- •31.4.3 ERPs
- •31.4.4 Further Measures
- •31.5.1 Rationale
- •31.5.2 Procedure
- •31.6 Closed-Loop Systems
- •31.7 Technical Requirements
- •31.8 Conclusion
- •References
- •32.1 Introduction
- •32.2.1 Equipment

212 Y. F. Low and R. Martinez-Cancino
17.1 Introduction
The two main objectives of preprocessing are:
. Data transformation: Raw EEG data is often unsuitable for analysis in its original
form. Preprocessing transforms it into a format that is appropriate for further
computations.
. Artifact handling: EEG signals are frequently contaminated with non-neural
signals,
artifacts need to be removed or minimized as much as possible to obtain clearer
and more accurate neural signals.
The following sections outline common preprocessing procedures; however, the
order in
they should be applied.
such as eye movements, muscle activity, or electrical noise. These
which they are presented does not necessarily reflect the sequence in which
17.2 Common Preprocessing Steps: Data Transformation
17.2.1 Inspecting Data
Once the data is loaded, conducting a quick inspection of its appearance and
properties is often beneficial. Things you shoul d or can do are as follows:
. Ensure that all EEG channels and their corresponding coordinates are correc tly
imported.
and for generating meaningful visualizations. For a detailed discussion on electrode placement and coordinate systems, refer to Robert Oostenveld’s blog on the
topic (Oostenveld, n.d.).
. Inspect the data for anomalies, such as missing event markers, incorrect marker
timings,
reference them with the experimental design or recording setup to identify
potential causes.
. Standardize event markers by renaming them for clarity and consistency.
Remove anythi
Accurate channe l locations are essential for various preprocessing steps
or flat (non-functioning) channels. If any irregularities are found, cross-
ng that is unnecessary for your analysis.
17.2.2 Changing the Sampling Frequency
Adjusting the sampling frequency involves either decreasing (downsampling) or
increasing (upsampling) the number of data points recorded per second. When data
are collected at a high sampling rate (e.g., >1000 Hz), it is often beneficial to
downsample the data offline to a lower frequency that is more suitable for analysis.
This can significantly reduce file size and improve processing efficiency.

17 EEG Preprocessing and Artifact Handling 213
In contrast, upsampling is less commonly used but can be helpful in specific
situations, for example to improve temporal precision when estimating ERP peak
latencies. Although it does not add new information, it allows more accurate
estimation of the timing of waveform peaks.
For analyses focused on frequency ranges up to the Beta band (approximately
30 Hz),
certain preprocessing steps, such as EMG analysis, fMRI gradient artifact correction,
or TMS pulse artifact removal may require higher sampling rates. After these steps
are compl eted, the data can usually be downsampled without loss of relevant
information.
aliasing. While
functions, which may lead users to overlook its importance.
a sampling rate between 150 and 300 Hz is typically sufficient. However,
Importantly, low-pass filtering must be applied before downsampling to avoid
this step is critical, it is often handled automatically within built-in
17.2.3 Re-referencing
Re-referencing implies altering the “baseline” of the recorded data offline using
different schemes to suit various purposes. Every re-reference scheme possesses
distinct benefits and limitations (Luck, 2014b; Yao et al., 2019). Generally,
re-referencing may be required for any of the following reasons:
. Impact on signal amplitude: If the online reference is near the channel of interest,
affect the signal’s amplitude, thereby reducing the desired effect. In this
it can
case, re-referencing the data to a channel not too close to the region of interest is
recommended.
. Artifact reduction: If all scalp channels exhibit
drifts), re-referencing helps mitigate these spatially widespread signals. The
Average Reference scheme, in particular, improves signal-to-noise ratio
(Bigdely-Shamlo et al.,
cient channel
channels
mobile EEG studies (Arad et al., 2018; Klug & Gramann, 2020).
. Hemispheric bias: EEG recordings often utilize a single electrode as the online
reference.
resulting voltage distribution may exhibit hemispheric asymmetry.
To mitigate this bias, re-referencing to the average of the left and right
mastoids
Reference scheme, which is most effective with high-density montages (typically
64 or more electrodes) (Nunez & Srinivasan, 2006) and at least 50% uniform
scalp coverag
. Comparability with
dependent. So, it would make sense to cross-check with other published research
papers in your area of research to know the reference they used. To facilitate
s (>64 channels) for even head coverage (Hu et al., 2018) and no bad
are included. This makes the Average Reference a common choice in
When referencing is limited to either the left or right mastoid, the
is recommended. An alternative is the previously mentioned Average
e (Luck, 2014c).
2015; Tsuchimoto et al., 2021), provided there are suffi-
literature: The choice of reference site can also be field-
common artifacts (e.g., line noise,

214 Y. F. Low and R. Martinez-Cancino
result comparison, it may be necessary to re-reference the data using the
referencing scheme employed in the literature being compared.
It is important to ensure that the new reference site is clean; otherwise, noise will
propagate
opportunity to explore different schemes and choose the most suitable one for a
specific research question or hypothesis.
to other channels after referencing. Overall, re-referencing provides the
17.2.4 Interpola ting Channels or Data Portions
Channel interpolation refers to estimating the signal at specific electrode locations
based on data from surrounding channels. This technique is commonly used to
replace EEG channels that are affected by persistent artifacts.
Many EEG processing tools use spherical spline interpolation by default, as it
generally
included in the interpolation process. The specific channels used often depend on the
chosen algorithm, user-defined parameters, and the characteristics of the dataset.
that can
when necessary. Overuse of interpolation can potentially distort the true underlying
brain activity. As such, interpretations based on interpolated data should be made
with caution.
mitigate
data portion within the pulse time window before proceeding with further
preprocessing or analysis. For more detailed strategies on managing EEG-TMS
data, see Hernandez-Pavon et al. (Hernandez-Pavon et al.,
yields high-quality results. However, not all scalp channels are necess arily
While there are no strict guidelines regarding the maximum number of channels
be interpolated, it is recommended to apply interpolation sparingly and only
In specific applications, such as EEG combined with TMS, interpolation can help
artifacts caused by the TMS pulse. A common approach is to interpolate the
2023).
17.2.5 Segmenting Data
Segmentation involves breaking down continuous EEG data into shorter, manageable segments or epochs. In studies focusing on event-related changes, continuous
data is divided into segments around events, defined by pre- and post-stimulus
intervals. These intervals should be selected based on the experimental paradigm
and the expected effects. Typical steps for generating ERPs are discussed in
Chap.
19: Event-Related Potentials.
Other types of analyses require specific segmentation considerations:
. Wavelet analysis:
left and right borders to avoid border and smearing effects (Herrmann et al., 2014;
Roach &
Mathalon, 2008).
The segment length should include a safety margin at both the

17 EEG Preprocessing and Artifact Handling 215
. FFT or Welch’s method: Data are divided into short, non-overlapping or
overlapping segments, which should be long enough to contain at least one
cycle of the lowest frequency of interest. Moreover, for computational efficiency, the number of data points per segment is often chosen as a power of two.
To learn more about spectral analysis, you are referred to Chap. 18: Introduction
to EEG
Oscillations and Spectral Analysis.
17.3 Common Preprocessing Steps: Artifact Handling
In Chap. 15: Getting Clean EEG Data, strategies to avoid or minimize artifacts
during data acquisition are presented. Despite these efforts, recorded data may still
be contaminated with various artifacts and noise. It is essential to remove or
minimize these artifacts to obtain neural signals that more accurat ely reflect the
underlying brain activity.
17.3.1 Filtering
Offline filtering attempts to modify the frequency characteristics of data, aimed at
reducing unwanted frequencies by removing noise from the data. Non-causal zerophase filters are often preferred because they preserve the phase of all frequency
components that remain in the signal, which is important for many analyses. Filters
generally fall into four categories:
. High-pass filter: Attenuates frequencies below the low-cutoff.
in EEG are between 0.1 and 0.5 Hz to reduce drifts such as body sway or skin
potentials.
. Low-pass filter: Reduces frequencies above the high-cutoff.
quencies that are usually studied lie below 40 Hz, this is a common cutoff
frequency. This helps min imize the impact of muscular artifacts and other highfrequency noise.
. Band-pass filter: The combination of
. Notch filter: A band rejec
its harmonics.
Furthermore, in
finite impulse response (FIR) and infinite impulse response (IIR), each based on
different design principles, but both are suitable for filtering EEG data. Note that, a
higher filter order produces a steeper magnitude response, resulting in stronger
suppression beyond the cutoff frequency and thus improved frequency selectivity.
However, this increased selectivity comes at the expenses of reduced temporal
precision.
publications, you might encounter two distinct types of filters:
tion filter used to attenuate line noise at 50 or 60 Hz and
high and low pass filters.
Common cutoffs
As most EEG fre-

216 Y. F. Low and R. Martinez-Cancino
Fig. 17.1 Magnitude
response plot
IIR Butterworth low-pass
filter, based on the halfpower cutoff definition.
About 70.7% of the original
amplitude is retained at the
cutoff frequency
of a 4th-order
When discussing cutoff frequencies, the two commonly used terms are half-
power cutoff and half-amplitude cutoff. The half-power cutoff, also known as the
3 dB point, refers to the frequency at which the power of the signal is reduced by
half (approximately 70.7% of the original amplitude), as illustrated in Fig.
contrast, the half-amplitude cutoff is defined by a
6 dB reduction, which corre-
17.1. In
sponds to 50% of the original ampl itude.
Filtering is a widely reported preprocessing step in publications. However, it is
advisable
to minimize the extent of filtering and apply it only, when necessary, as
filtering can unpredictably alter the data. Interpreting the results with caution is
essential. This topic has been thoroughly discussed, and practical guidelines for
applying filters to EEG data are published (de Cheveigne & Nelken, 2019; Widmann
et al., 2015; Zhang et al., 2024a, b).
17.3.2 Attenuating Artifacts
Stereotypical artifacts such as blinks, lateral eye movements, and muscle activity
often exhibit consistent shapes and distributions. They can be reduced or removed
through correction methods without losing data points.
17.3.2.1 Independent Component Analysis (ICA)
ICA is a blind source separation (BSS) method which involves decomposing the
recorded
These artifacts have characteristic topographies, shapes, and time courses, such as
blinks and eye movements, as illustrated in Fig.
after remo
tively reduced.
(Makeig et al., 1996), Fast ICA (Hyvärinen & Oja, 2000), RELICA (Artoni et al.,
2014), and AMICA (Klug et al., 2024; Palmer et al., 2010). The choice of the ICA
algorithm
EEG signal into components that represent brain activity and artifacts.
17.2. By reconstructing the signal
ving the artifact-related components, the influence of artifacts is effec-
Various ICA-based
methods have been proposed, including Infomax ICA
depends on the specific characteristics of your data and the type of artifacts

17 EEG Preprocessing and Artifact Handling 217
Fig. 17.2 Topographic maps of blinks and eye movements. (a) Blinks are prominent bell-shaped
deflections over the data. They show a predominantly frontal topography. (b) Lateral eye movements show a box-shaped morphology, appear strongest in frontopolar and frontotemporal channels, and typically show opposite polarity on each side of the scalp. Note: the interval between the
vertical green lines is 1 s
you are dealing with (Delorme et al., 2012; Dimigen, 2020; Frølich & Dowding,
2018; Gorjan et al., 2022; Hoffmann, 2009; Klug et al., 2024; Leutheuser et al.,
2013; Stergiadis et al., 2022). Among these, Infomax ICA is perhaps the most used
and freque
ntly cited in publications.
The quality of ICA can be affected by several factors, which should be carefully
considered:
. Number of EEG channels: The maximum number of components that can be
isolated in
EEG data corresponds to the number of electrodes used. While
decomposition can be performed with low-density recordings, it is generally
recommended to use at least 64 EEG channels for optimal results (Cohen,
2014; Klug & Gramann, 2020). Increasing the number of channels enhances
spatial resolution
. Amount of trai
and improves the accuracy of independent component analysis.
ning data: A minimum of (number of channel)^2
20 (for fewer
than 64 channels) or (number of channel)^2 30 (for more than 64 channels)
data points is recommended to perform ICA (Makeig & Onton, 2011). It is also
important
that the training data contain sufficient representative neuronal activity,
including artifacts intended to be remove.

218 Y. F. Low and R. Martinez-Cancino
. Quality of training data: Slow drifts in the data can significantly deteriorate ICA
performance. To mitigate this, applying a high-pass filter is strongly
recommended. The cutoff frequency should be selected in a balanced way that
is neither too low nor too high (i.e., usually not more than 1 Hz). Some literature
suggests a cutoff between 1 and 2 Hz for clean decompositions, while continuing
the analyses on less or unfiltered data afterwards to avoid distorting the time
course of ERPs (Debener et al.,
2010; Klug & Gramann, 2020; Winkler et al.,
2015). Additionally, applying a low-pass filter reduces high-frequency noise,
preventing it from dominating ICA components and improving the extraction
of meaningful signals.
Furthermore, the following points should be kept in mind when applying ICA:
. Over-correction: ICA is purely data-driven, meaning that the decomposed com-
ponents
are based on statistical properties of the signal and do not map precisely
onto physiological processes. When using ICA for artifact removal, it is important to carefully examine all decomposed components to make well-informed
decisions about whether a component should be discarded. Over-correction,
which occurs when the components removed to correct artifacts also contain
valid EEG, should be avoided.
. Rank-deficiency: ICA assumes that the signal
is a linear mixture of independent
sources, with the maximum number of extractable source signals limited by the
number of channels. However, when channels are interpolated or re-referenced to
a common average (or mean of mastoids), the principle of independence is
violated. This results in rank-deficiency, which must be addressed to ensure a
proper computation of the inverse ICA weight matrix. Mitigation of this issue
involves running ICA with a reduced number of components or performing ICA
before interpolation or re-referencing.
17.3.2.2 Regression Techniques
Regression-based methods, such as the Gratton and Coles algorithm (Gratton et al.,
1983), can be utilized for the identification and correction of ocular artifacts. These
methods
leverage the EOG channels to estimate the influence of blinks and eye
movements on the EEG channels via regression. Ocular artifacts are then subtracted
based on a calculated correction factor. This approach is computationally efficient,
though optimal performance necessitates EOG recordings. For a comprehensive
overview of current methods for handling ocular artifacts, see (Jiang et al.,
et al.,
Ronca
2024).
2019;
Another notable method is the Carbon Wire Loop (CWL) regression technique,
desig
ned to reduce motion artifacts in brain signal recordings, particularly in simultaneous EEG-fMRI studies (van der Meer et al., 2016). CWLs are flexible carbon
wires placed
near the electrodes that capture motion-related signals without detecting
neural activity. These motion signals are used as regressors in a linear model to

17 EEG Preprocessing and Artifact Handling 219
estimate and remove motion-induced noise from the recorded brain data, significantly enhancing signal quality.
17.3.2.3 Template Subtraction Methods
Methods such as Average Artifact Subtraction (AAS) (Allen et al., 2000) can be used
to remov
environment. In this method, artifact templates are calculated by averaging over
adjacent artifact epochs and subsequently the templates are subtracted to reduce the
artifacts. More details on handling artifacts for EEG-fMRI data can be found in
Warbrick’s review paper (Warbrick, 2022).
e gradient and cardioballistic artifacts in EEG data collected within an MR
17.3.3 Rejecting Artifacts
Non-stereotypical artifacts can sporadically appear on various electrodes and exhibit
different shapes. Examples include sudden electrode pops across multiple channels,
significant movements by the participant affecting several channels, or transient
muscular tensions impacting different regions of the head. These artifacts can be
carefully addressed through interpolation or completely rejected. Rejection involves
removing portions of the affected data or specific channels, or in extreme cases,
discarding the entire dataset from a participant, to prevent compromising the group
analysis and study results.
Artifacts must be detected before they can be rejected. This can be achieved either
manually
for automatic identification. There are no standardized parameters for artifact detection as it varies depending on the unique characteristics of the data in each experiment. Therefore, it’s crucial to identify parameters that best suit the current dataset.
Ideally, these parameters should be consistently applicable to all datasets within a
study. However, exceptions may arise; for instance, if some datasets exhibit higher
levels of noise, it may be sensible to adjust the criteria accordingly.
ing attenuation methods could lead to an unrepresentative sample of trials, particularly in mobile environments. Therefore, rather than using only one approach, a more
effective strategy is to combine correction with rejection (Zhang et al., 2024c). As a
general pract
able patterns, such as blinks.
by using expert judgment to mark the data, or by applyi ng specific criteria
It is important to note that relying solely on artifact rejection without incorporat-
ice, give priority to removing sporadic artifacts and correcting predict-

220 Y. F. Low and R. Martinez-Cancino
17.4 What to Consider When Developing a Preprocessing
Pipeline?
The recent recommendation to either avoid or minimally preprocess EEG data, based
on the assertion that most preprocessing techniques do not improve data quality
(Delorme, 2023), has sparked controversy. This view has been challenged by several
EEG expert
argue that while such an approach may b e applicable to relatively clean EEG
recordings, preprocessing can be crucial for improving the quality and interpretability of noisier or more complex EEG data.
Therefore, it is generally imprudent to completely disregard data preprocessing.
To suppor
preprocessing pipeline for a specific dataset, several key factors have been outlined
for consideration:
. Data characteristics: Not all preprocessing steps are always relevant; their
necessity
demand more comprehensive preprocessing and artifact rejection. Consequently,
this may involve interpolating more channels, applying more stringent filtering,
and adjusting the criteria for artifact d etection.
. Intended analysis and research goals: The desired outcome measure can dictate
the nec
Laplacian filter (also known as Current Source Density, CSD) as a preprocessing
step before performing connectivity analysis can mitigate the effect of volume
conduction through the scalp, leading to a more precise calculation of connectiv-
ity. If the research objective is to compare your findings with a previous study, it
would make sense to replicate the processing pipeline reported in that publication,
rather than creating a different one, to ensure comparability.
. Knowledge of preprocessing steps: Understanding the prerequisites for specific
analysis
noise, as well as rejecting non-stereotypical artifacts, can enhance the effective-
ness of ICA. Additionally, interpolating bad channels prior to applying an
average reference helps to maintain a balanced spatial representation across the
scalp, which is important for preserving the accuracy of topographical and
source-level analyses.
. Suitability: Some steps are better suited for conti nuous or for segmented data,
while others can be applied to both, depending on requirements. For example,
editing the markers or re-referencing can be performed on both segmented and
continuous data. However, high-pass filtering is best applied to long continuous
data to avoid edge artifacts that might contaminate the region of interest on
shorter segments. Similarly, ICA is more efficient when applied to continuous
data, as larger, uninterrupted data allow for better separation of components.
. Order: Some steps within a pipeline can be interchangeable. However, others can
significantly influence the final output, depending on whether they are executed
s (e.g., de Cheveigné, 2023; Larsen & Versace, 2024). These experts
t informed decision-making in designing a valid and effective EEG
depends on the quality of the acquired data. Noisy datasets often
essary processing steps and their sequence. For example, using a
steps is crucial. For instance, attenuating slow drifts and high-frequency

17 EEG Preprocessing and Artifact Handling 221
earlier or later and whether they are linear or non-linear operations. Luck com-
prehensively explains these concepts in his book and also provides practical
guidance for developing processing pipelines for ERP analysis (Luck,
Equally
documented in detail to ensure proper evaluation and reproducibility (Keil
et al., 2014).
interconnected
fully plan and refine the workflow will help ensure reliable and interpretable
results.
important are the parameters used at each step, which should be
As you can see, constructing an EEG processing pipeline involves many
decisions. Therefore, dedicating sufficient time to thought-
2014a).
17.5 Tools for Processing and Analyzing EEG
A variety of software can be utilized to process and analyze EEG data, encompassing
both open-source and proprietary options. The choice of software largely depends on
the research project’s requirements and the user’s programming proficiency. Finding
a single application that meets all needs is often challenging. Rather than focusing on
one particular tool, it may be more effective to adopt a platform that can interact
with multiple software to some extend and allow, where possible, the creation of
custom scripts. Below is a list of some software options (the order is arbitrary
and the list is not exhaustive). Some other tools are mentioned in the review paper
by Das et al. (Das et al.,
2023):
. EEGLAB (Delorme & Makeig, 2004): An open-source MATLAB toolbox offering
a comprehensi
preprocessing, visualization, and analyzing ERPs and time-frequency data. Sev-
eral automated pipelines have been developed based on this platform, such as the
PREP (Bigdely-Shamlo et al.,
Monachino
ADJUST (Mogn
. ERPLAB Toolbox: A free, open source Matlab package for analyzing ERP data,
closely
. BrainVision Analyzer: A commercial software package that provides tools for the
analysis
for its user-friendly interface, intuitive features, and reliable scientific methods. It
also allows interfacing with MATLAB and extending processing functionalities
via macro scripting.
. FieldTrip: Another open-source MAT
EEG, and other neurophysiological data. It offers advanced functions for time-
frequency analysis, source reconstruction, and statistical analysis.
. CURRY: A
and neurophysiological data analysis. It’s widely used in both research and
clinical settings, particularly for epilepsy evaluation.
ve environment for processing EEG data, including functions for
2015), HAPPE (Gabard-Durnam et al., 2018;
et al.,
2022), RELAX (Bailey et al., 2023a, b; Hill et al., 2024),
on et al., 2011), and DISCOVER-EEG (Gil Ávila et al., 2023).
intergrated with EEGLAB (Lopez-Calderon & Luck, 2014).
of EEG, ERP, and other neurophysiological data. The software is known
LAB toolbox designed for analyzing MEG,
commercial software offering tools for multi-modal neuroimaging
Соседние файлы в папке Библиотека им академика М.И. Перельмана
