EEG Noise and Artifacts

Introduction to EEG Noise

Noise in EEG is defined as any activity that isn't brain activity. The lecture focuses on identifying, dealing with, and eliminating noise to analyze EEG signals effectively.

Exam Structure

The exam has two sections, A and B, each with multipart questions.

• Section A: Offers a choice of three multipart questions.

• Section B: Contains only one multipart question, which everyone must answer.

• The exam includes writing directly on the paper to limit excessive writing.

Exam Content

Questions will be based on the first lecture, this lecture, and the next one. Practical lectures are excluded.

Support Class

A two-hour support class is scheduled after the class test to address exam-related questions on the covered material.

Sources of Noise in EEG Data

Eye Movements and Blinking

• Eye movements create a significant signal due to the dipole nature of the eye.

• Blinking produces a battery effect as tears interact with the eyeball and eyelid, generating electrical signals.

Muscle Activity

• Biting: Muscle tension from biting can produce substantial noise, especially when subjects are asked to concentrate and keep still.

• Neck Tension: Chin rests can cause discomfort, leading to neck tension and measurable artifacts.

• EMG (myography) measures muscle activity.

Heartbeats

Heartbeats can be picked up by EEG electrodes, but their impact varies depending on the frequency band being analyzed.

Skin Potentials

Inconsistent skin impedance during electrode placement can lead to electron flow and skin potentials, distorting measurements.

Movement

• Any movement, even without muscle involvement, changes the electrical environment around EEG electrodes.

• Electromagnetic fields in the environment can interact with moving electrodes, adding noise.

EEG electrodes signal vulnerability

EEG electrode signal is very sensible to movement through space, that's one of the big issues with using EEG in real world situations.

Examples of Noisy EEG Measurements

Horizontal Eye Movements

Produce boxcar-shaped artifacts in EEG signals, typically captured by dedicated horizontal electrooculography (HEOG) channels.

Vertical Eye Movements and Blinks

Generate pervasive artifacts, visible across many channels, especially in the vertical electrooculography (VEOG) signal.

Teeth Gritting and Heartbeats

Can also be identified, although heartbeat artifacts are generally smaller when measured from scalp electrodes.

Methods for Dealing with Artifacts

Manual Artifact Rejection

• Involves visually inspecting raw EEG data and manually flagging artifact segments.

• Experimenters click on the beginning and the end of the blink to identify to the software that there was a blink activity at that time.

• Flagged data is typically discarded, either partially or entirely at the trial level.

• It's a slow method.

• Useful for spotting distinctive artifacts that are hard to automate.

Automated Threshold-Based Rejection

• Uses predefined voltage thresholds to automatically detect and reject trials containing artifacts.

• Commonly uses VEOG signals to detect blinks, discarding any trial where the signal exceeds a set threshold.

• 60 millivolts is the standard number.

• It's a very fast method.

Averaging in ERP Research

• After artifact rejection, the remaining data is averaged to enhance the signal-to-noise ratio.

• The noise remaining is proportional to the inverse of the square root of the number of trials: $Noise</p></li></ul><p>\propto \frac{1}{\sqrt{n}}$

  • Systematic artifacts need to still be rejected because they won't cancel out during averaging.

  Robust Averaging

  • Uses the median to diminish the role of outliers.

  • Weights data points based on their distance from the median, reducing the influence of extreme values while retaining all data.

  • Things that are far away from the median contribute still to the median but less to a lesser extent.

  • Computationally intensive.

  There's software who does that.

  Benefits of Roboust Averaging

  Great data preservation, you don't have to throw any data away with this method.

  EOG Artifact Correction

  *   Targets eye movement artifacts by measuring individual blink characteristics using dedicated EOG electrodes.
  *   EOG signal is substrated from the real EEG.
  *   A real EEG is always going to be whatever we measured minus artifacts.
  *   Employs methods like the Croft and Barry technique to account for different artifact sources.
  

  $EEG = Measured - Artifacts$

  $Pure EEG = Measured - EOG$

  Principal Component Analysis (PCA)

  The goal is to change the numbers we have into numbers that are more useful to us.

  • It rotates the data axes to align with the directions of maximum variance, creating orthogonal components.

  • It Finds a set of orthogonal components (at right angles each other).

  • Components correlating with known artifacts, like EOG, are removed.

  • It Assumes greatest variation in EEG data is noise.

  Independent Component Analysis (ICA)

  Each EEG electrode measures voltage from multiple sources, it tries to separate all those sources.

  The principle behind it is completely sound.

  • It Starts with the principle that the thing you're measuring is a mixture of measurements of separate sources.

  • Employs the central limit theorem in reverse, aiming to unmix sources by finding components with non-normal distributions.

  • Requires the sources to be statistically independent, and the original sources should preferably be non-normally distributed.

  • The standard normal ICA form has no notion of scale, no notion of sign and no notion of order.

  Conclusion

  The lecture detailed various methods for identifying and mitigating noise in EEG data, ranging from manual techniques to advanced analytical approaches like PCA and ICA. Each method has its strengths and limitations, influencing their suitability for different research contexts.

  Introduction to EEG Noise

  Noise in EEG is defined as any activity that isn't brain activity, encompassing a wide range of non-cerebral signals that contaminate EEG recordings. The primary concern is to distinguish genuine neural oscillations from artifacts to accurately analyze brain activity. Understanding and mitigating these noises are crucial for drawing valid conclusions from EEG data. The lecture focuses on identifying, dealing with, and eliminating noise to analyze EEG signals effectively.

  Exam Structure

  The exam has two sections, A and B, each with multipart questions.

  • Section A: Offers a choice of three multipart questions. This allows students to select questions that align with their strengths and areas of focus.

  • Section B: Contains only one multipart question, which everyone must answer. This question is designed to assess core knowledge and understanding of key concepts covered in the lectures.

  • The exam includes writing directly on the paper to limit excessive writing. This format helps in managing time effectively and focusing on concise, relevant answers.

  Exam Content

  Questions will be based on the first lecture, this lecture, and the next one. Practical lectures are excluded. The content will cover theoretical concepts, noise identification, and artifact correction methods discussed in these lectures.

  Support Class

  A two-hour support class is scheduled after the class test to address exam-related questions on the covered material. This session provides an opportunity for students to clarify doubts and reinforce their understanding before the exam.

  Sources of Noise in EEG Data

  Eye Movements and Blinking

  • Eye movements create a significant signal due to the dipole nature of the eye. These movements generate large amplitude deflections in the EEG signal, particularly in frontal electrodes.

  • Blinking produces a battery effect as tears interact with the eyeball and eyelid, generating electrical signals. This effect results in a broad, high-amplitude artifact that can obscure underlying EEG activity.

  Muscle Activity

  • Biting: Muscle tension from biting can produce substantial noise, especially when subjects are asked to concentrate and keep still. The temporalis and masseter muscles, involved in biting, can introduce high-frequency noise into the EEG signal.

  • Neck Tension: Chin rests can cause discomfort, leading to neck tension and measurable artifacts. The trapezius and sternocleidomastoid muscles are common sources of these artifacts.

  • EMG (myography) measures muscle activity. This technique is often used in conjunction with EEG to monitor and identify muscle-related artifacts.

  Heartbeats

  Heartbeats can be picked up by EEG electrodes, but their impact varies depending on the frequency band being analyzed. Cardiac artifacts are typically more pronounced in lower frequency bands and can be mistaken for slow brain oscillations.

  Skin Potentials

  Inconsistent skin impedance during electrode placement can lead to electron flow and skin potentials, distorting measurements. Ensuring proper skin preparation and electrode application is crucial to minimize these artifacts.

  Movement

  • Any movement, even without muscle involvement, changes the electrical environment around EEG electrodes. This includes small movements like fidgeting or head motion.

  • Electromagnetic fields in the environment can interact with moving electrodes, adding noise. Sources include power lines, electronic devices, and other equipment in the vicinity.

  EEG electrodes signal vulnerability

  EEG electrode signal is very sensible to movement through space, that's one of the big issues with using EEG in real world situations. Even slight movements can introduce artifacts, making it challenging to collect clean EEG data outside controlled laboratory settings.

  Examples of Noisy EEG Measurements

  Horizontal Eye Movements

  Produce boxcar-shaped artifacts in EEG signals, typically captured by dedicated horizontal electrooculography (HEOG) channels. These artifacts appear as rectangular-shaped waves in the EEG, making them relatively easy to identify.

  Vertical Eye Movements and Blinks

  Generate pervasive artifacts, visible across many channels, especially in the vertical electrooculography (VEOG) signal. Blinks create large, sharp deflections that can contaminate a significant portion of the EEG data.

  Teeth Gritting and Heartbeats

  Can also be identified, although heartbeat artifacts are generally smaller when measured from scalp electrodes. Teeth gritting produces high-frequency, irregular noise, while heartbeat artifacts appear as rhythmic deflections synchronized with the pulse.

  Methods for Dealing with Artifacts

  Manual Artifact Rejection

  • Involves visually inspecting raw EEG data and manually flagging artifact segments. This requires trained personnel to identify and mark segments of data containing artifacts.

  • Experimenters click on the beginning and the end of the blink to identify to the software that there was a blink activity at that time. This ensures the software accurately excludes the contaminated data.

  • Flagged data is typically discarded, either partially or entirely at the trial level. The decision to discard data depends on the severity and extent of the artifacts.

  • It's a slow method. Manual artifact rejection is time-consuming and labor-intensive.

  • Useful for spotting distinctive artifacts that are hard to automate. This method is particularly effective for identifying novel or unusual artifacts that automated algorithms may miss.

  Automated Threshold-Based Rejection

  • Uses predefined voltage thresholds to automatically detect and reject trials containing artifacts. This method relies on setting specific voltage limits beyond which data is considered artifactual.

  • Commonly uses VEOG signals to detect blinks, discarding any trial where the signal exceeds a set threshold. Blinks are identified by their large amplitude in the VEOG channel.

  • 60 millivolts is the standard number. A threshold of 60 $mV$ is commonly used to detect blinks, but this value may vary depending on the specific EEG system and recording conditions.

  • It's a very fast method. Automated threshold-based rejection can quickly process large amounts of data.

  Averaging in ERP Research

  • After artifact rejection, the remaining data is averaged to enhance the signal-to-noise ratio. Averaging reduces the impact of random noise, making underlying ERP components more visible.

  • The noise remaining is proportional to the inverse of the square root of the number of trials: $Noise \propto \frac{1}{\sqrt{n}}$

  • Systematic artifacts need to still be rejected because they won't cancel out during averaging. Systematic artifacts, such as consistent muscle activity or line noise, will persist through averaging and must be removed.

  Robust Averaging

  • Uses the median to diminish the role of outliers. Robust averaging is less sensitive to extreme values than traditional averaging methods.

  • Weights data points based on their distance from the median, reducing the influence of extreme values while retaining all data. This weighting ensures that outliers have a minimal impact on the final averaged signal.

  • Things that are far away from the median contribute still to the median but less to a lesser extent. This approach ensures that all data is used, but the influence of noisy data is minimized.

  • Computationally intensive. Robust averaging requires more processing power than simple averaging.

  There's software who does that.

  Benefits of Roboust Averaging

  Great data preservation, you don't have to throw any data away with this method. This is particularly useful when dealing with datasets that have a limited number of trials.

  EOG Artifact Correction

  *   Targets eye movement artifacts by measuring individual blink characteristics using dedicated EOG electrodes.
  *   EOG signal is substrated from the real EEG.
  *   A real EEG is always going to be whatever we measured minus artifacts.
  *   Employs methods like the Croft and Barry technique to account for different artifact sources.
  

  $EEG = Measured - Artifacts$

  $Pure EEG = Measured - EOG$

  Principal Component Analysis (PCA)

  The goal is to change the numbers we have into numbers that are more useful to us.

  • It rotates the data axes to align with the directions of maximum variance, creating orthogonal components. PCA transforms the data into a new coordinate system where the principal components represent the directions of maximum variance.

  • It Finds a set of orthogonal components (at right angles each other). These components are uncorrelated, simplifying the analysis and interpretation of the data.

  • Components correlating with known artifacts, like EOG, are removed. By identifying and removing artifact-related components, the EEG signal can be cleaned.

  • It Assumes greatest variation in EEG data is noise. This assumption is based on the idea that artifacts often have larger amplitudes than neural signals.

  Independent Component Analysis (ICA)

  Each EEG electrode measures voltage from multiple sources, it tries to separate all those sources.

  The principle behind it is completely sound.

  • It Starts with the principle that the thing you're measuring is a mixture of measurements of separate sources. ICA aims to decompose the EEG signal into a set of independent components, each representing a distinct source.

  • Employs the central limit theorem in reverse, aiming to unmix sources by finding components with non-normal distributions. This approach is based on the assumption that the underlying sources are non-Gaussian.

  • Requires the sources to be statistically independent, and the original sources should preferably be non-normally distributed. The effectiveness of ICA depends on the degree to which these assumptions are met.

  • The standard normal ICA form has no notion of scale, no notion of sign and no notion of order. This means that the components are not inherently ordered or scaled, requiring further interpretation.

  Conclusion

  The lecture detailed various methods for identifying and mitigating noise in EEG data, ranging from manual techniques to advanced analytical approaches like PCA and ICA. Each method has its strengths and limitations, influencing their suitability for different research contexts. Understanding these methods is crucial for conducting reliable EEG research.

  Here are detailed answers to the exam questions:

  1. What is the definition of noise in EEG data, and why is it important to mitigate?

     • Noise in EEG data is defined as any activity that isn't brain activity. This includes a wide range of non-cerebral signals that contaminate EEG recordings.

     • It is important to mitigate noise to accurately analyze brain activity and draw valid conclusions from EEG data. Distinguishing genuine neural oscillations from artifacts is crucial for reliable research.

  2. Describe the structure of the exam, including the number of sections and types of questions in each section.

     • The exam has two sections: A and B.

     • Section A offers a choice of three multipart questions, allowing students to select questions that align with their strengths.

     • Section B contains only one multipart question, which everyone must answer. This question assesses core knowledge and understanding of key concepts.

     • The exam includes writing directly on the paper to limit excessive writing and manage time effectively.

  3. What topics will be covered in the exam, and which lectures should students focus on for preparation?

     • The exam covers topics discussed in the first lecture, the current lecture on EEG noise, and the subsequent lecture.

     • Practical lectures are excluded.

     • Students should focus on theoretical concepts, noise identification, and artifact correction methods discussed in these lectures.

  4. Explain the dipole nature of eye movements and how they create artifacts in EEG signals.

     • Eye movements create a significant signal due to the dipole nature of the eye, where the cornea is positively charged relative to the retina.

     • These movements generate large amplitude deflections in the EEG signal, particularly in frontal electrodes, leading to artifacts.

  5. How does blinking produce a battery effect, and what type of artifact does it generate?

     • Blinking produces a battery effect as tears interact with the eyeball and eyelid, generating electrical signals.

     • This effect results in a broad, high-amplitude artifact that can obscure underlying EEG activity.

  6. Describe how muscle activity, such as biting or neck tension, can introduce noise into EEG recordings.

     • Muscle activity, such as biting, can produce substantial noise due to the tension in muscles like the temporalis and masseter.

     • Neck tension, often resulting from discomfort with chin rests, can also introduce measurable artifacts from muscles like the trapezius and sternocleidomastoid.

     • These muscle activities introduce high-frequency noise into the EEG signal.

  7. Why is consistent skin impedance important for EEG measurements, and how can inconsistent impedance distort measurements?

     • Consistent skin impedance is important for EEG measurements because it ensures proper electrical contact between the electrodes and the scalp.

     • Inconsistent skin impedance can lead to electron flow and skin potentials, distorting measurements and introducing artifacts.

     • Ensuring proper skin preparation and electrode application is crucial to minimize these artifacts.

  8. How do electromagnetic fields in the environment interact with EEG electrodes to add noise?

     • Electromagnetic fields in the environment, from sources like power lines and electronic devices, can interact with moving EEG electrodes, adding noise.

     • Even slight movements can introduce artifacts, making it challenging to collect clean EEG data outside controlled laboratory settings.

  9. What are boxcar-shaped artifacts, and which type of movement typically produces them?

     • Boxcar-shaped artifacts are rectangular-shaped waves in the EEG signal.

     • They are typically produced by horizontal eye movements and are captured by dedicated horizontal electrooculography (HEOG) channels.

  10. Explain how manual artifact rejection is performed, and what are its advantages and disadvantages?

      • Manual artifact rejection involves visually inspecting raw EEG data and manually flagging artifact segments.

      • Experimenters click on the beginning and the end of the artifact to identify it to the software.

      • Advantages: Useful for spotting distinctive artifacts that are hard to automate; effective for identifying novel or unusual artifacts.

      • Disadvantages: Slow, time-consuming, and labor-intensive.

  11. How does automated threshold-based rejection work, and what are its typical settings for blink detection?

      • Automated threshold-based rejection uses predefined voltage thresholds to automatically detect and reject trials containing artifacts.

      • It commonly uses VEOG signals to detect blinks, discarding any trial where the signal exceeds a set threshold.

      • A threshold of 60 $mV$ is commonly used for blink detection, but this may vary.

  12. Describe the concept of averaging in ERP research and how it enhances the signal-to-noise ratio.

      • Averaging in ERP research involves averaging the remaining data after artifact rejection to enhance the signal-to-noise ratio.

      • Averaging reduces the impact of random noise, making underlying ERP components more visible.

      • The noise remaining is proportional to the inverse of the square root of the number of trials: $Noise \propto \frac{1}{\sqrt{n}}$

  13. What is robust averaging, and how does it diminish the role of outliers in EEG data?

      • Robust averaging uses the median to diminish the role of outliers, making it less sensitive to extreme values than traditional averaging methods.

      • It weights data points based on their distance from the median, reducing the influence of extreme values while retaining all data.

      • This approach ensures that all data is used, but the influence of noisy data is minimized.