Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

by Fernando Cross Villasana, Ph.D.
Scientific Consultant (Brain Products)

Logo Analyzer

Noise and artifacts are an unavoidable aspect of EEG recordings. As such, handling them is an invaluable skill for EEG researchers to make the best of their data. In this article we present an overview of common artifacts together with tools and strategies for managing various types of artifacts in BrainVision Analyzer 2.

For tips on the recording set up to avoid artifacts, stay tuned for the next issue of our newsletter.

Overview

1. Introduction

When recording EEG, the aim is to obtain a clear signal from the brain to help us investigate an aspect of its inner workings. However, intruding signals from other sources often enter the recording as well, obscuring the EEG signal of our interest. These intruding signals are known as “artifacts” and can have various physiological and non-physiological origins.

As EEG signals typically range in low amplitudes of tens of microvolts, they can be easily blurred by artifacts, reducing the signal to noise ratio. For example, activity from head muscles can overlap with oscillations from the brain, or movement of the cap creates distortions that affect the amplitude of an ERP. Technically speaking, artifacts add uncontrolled variability to the data, which confounds experimental observations. As such, even small artifacts can reduce statistical power in a study and alter results if they happen frequently.

Best efforts should always be made to prevent artifacts from entering EEG recordings. Nonetheless a certain level of artifact intrusion remains unavoidable, especially in the increasing field of mobile EEG and out-of-lab settings. But don’t worry, despite the unavoidability of artifacts, it is possible to obtain good quality signals with the right tools and artifact handling strategies.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 1: EEG data before (black) and after artifact handling (red) in BrainVision Analyzer 2.

2. Types of artifacts

Artifacts can be broadly classified into two groups according to their origin: physiological and technical. Physiological artifacts are signals generated by the human body which are not EEG. Technical artifacts, originate from the EEG equipment and from the environment.

As physiological artifacts tend to overlap and obstruct the EEG signal, data preprocessing often helps in separating them from the data. On the other side, while some technical artifacts also obstruct the signal, many others are an indicator that the EEG signal is lost altogether. For this reason, they are best prevented when setting up the recording. In the current article we focus on artifact handling during offline preprocessing. For a guide on artifact prevention during recording, stay tuned for the next issue of our newsletter where we will publish an article on recording settings.

2.1.   Physiological artifacts

Eye blink

The blink artifact is generated by the eye’s potential difference between the positively charged cornea and the negative retina: as the eyelid slides and the eyeball rotates during blinking the polarity is inverted and a positive current towards the scalp is created [1]. The artifact is most prominent at frontal channels close to the eyes, reaching over a hundred microvolts in amplitude. In the frequency domain, it contains frequencies mostly in the EEG delta and theta bands.

Blink artifacts decline with greater distance from the eyes but still affect posterior channels. Moreover, blinking momentarily alters visual processing in the brain, affecting the portion of EEG that follows the blink [2]. For this reason, if blinking happens systematically (e.g., right after each stimulus presentation), it can affect experimental observations [3].

→ Handling methods: ICA, regression-based subtraction.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 2: Blink artifacts over continuous data. The middle panel displays the spectral profile of the blink under the shaded area in a vertical EOG channel (VeogU) and its projection to a frontal and an occipital channel. The topography shows the typical distribution of blinks over the scalp.

Eye movements

Similar to blinks, lateral eye movements such as saccades generate a current away from the eyeballs, but this time towards the sides of the head [2], producing a box-shaped deflection with opposite polarity on each side. They are most prominent over channels close to the temples, but also affect channels outside these regions. In the frequency spectrum, the box shape created by eye movements peaks in the delta and theta bands, but has effects up to 20 Hz [4].

→ Handling methods: artifact rejection, ICA, regression-based subtraction.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 3: Eye movement artifacts over continuous data. The middle panel displays the spectral profile of the eye movement under the shaded area in a horizontal EOG channel (HeogL), and its projection to a frontal and a parietal channel. The topography shows the typical distribution of eye movements over the scalp.

Muscular artifacts

Muscular activation produces high frequencies that overlap with the entire EEG spectrum. Their amplitude depends on the intensity of activation and the muscle groups involved [5]. Most notably, activation during teeth clenching generates large noise that extends to the whole scalp. However, slighter artifacts are produced by other muscle groups of the head such as jaw or forehead [6, 7].

Shoulder and neck tension lead to a persistent artifact that reaches lower electrodes including the mastoid region. In this case, if mastoid channels are used for re-referencing, the muscular artifact is introduced into all other channels. In frequency space, muscular artifacts are most prominent above 20 Hz, and up to 300 Hz [5], however they cover a broad spectrum that affects all EEG frequencies [7].

→ Handling methods: Most commonly artifact rejection. Filtering can attenuate the impact. ICA can remove persistent localized muscle artifacts.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 4: A. Teeth clenching artifact in the time domain and frequency representation of selected frontal, temporal and parietal channels. B. Persistent neck tension artifact affecting channel TP10 close to the mastoids. When averaged TP9 and TP10 are used as a new reference, the artifact is introduced in all other channels.

Pulse artifact

Pulsation of the head arteries generated by the heartbeat can lead to a slight rhythmical movement of the electrodes [3]. In simultaneous EEG-fMRI recordings, the participant’s supine position and scanner environment magnify the artifact [8]. However, pulse artifacts also occur sporadically under normal laboratory conditions, in participants with hypertension, or related to physical activity.

The pulse artifact also affects the mastoid region due to its proximity to the carotid artery. In this case, re-referencing to the mastoids introduces the pulse to all channels [3]. Because of their small magnitude, pulses can easily be confused with EEG rhythms, introducing confounds in the time and frequency domains. A particular example lays in the field of epilepsy, where periodic pulse artifacts can be confounded with epileptiform activity [9].

→ Handling methods: Best implemented with co-registered electro-cardiogram (ECG), mandatory in EEG-fMRI. Special algorithms can then identify the heartbeat and remove the pulses. Without ECG, ICA or the average reference can limit the pulse’s influence.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 5: Pulse artifact spikes over an EEG recording. Notice how at some points the pulsations are indistinguishable from EEG signal. The topography of the highlighted peak for one pulsation is shown to the right. It reflects the most common distribution of pulse artifacts.

Sweating / skin potentials

Even mild sweating leads to changes in the conductivity of the skin. This produces a drifting voltage on the scalp that shows as slow drifts in the recording [10]. The drifts can vary widely in frequency and magnitude. This artifact is caused by warm environments, during physical activity, or during stress. The fluctuation induced by these drifts affects the timing and amplitude of signals in the time domain such as ERPs [10]. In frequency space, this artifact contains power mostly in slow frequencies.

→ Handling methods: High-pass filtering reduces drift. However, residual drift is hard to eliminate and it is better avoided by having a fresh and dry environment in the lab. Because of the signal fluctuation of the drift, a greater number of trials is required for averaging procedures to stabilize the signal.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 6: Sweat artifact producing slow drift over various EEG channels. The spectral profile shown for some channels reflects peak amplitude under 1 Hz.

Body movement

Movement of the body leads to slight displacement of the EEG cap over the scalp, especially when the cap is loose. This alters electrode impedance levels in the process, leading to artifacts. Gross movements produce large shifting voltages that can even saturate amplifiers momentarily [3]. Slight body sway leads to gradual drift in the channels. Complex movements in certain tasks produce equally complex movements of the cap involving pulling, sliding and shaking, which affect all channels [e.g. 11].

→ Handling methods: Sporadic movements may be discarded through artifact rejection, while gradual drift can be attenuated through filtering. Complex shaped artifacts can sometimes be subtracted via ICA.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 7: Drift on all channels caused by body sway.

2.2.   Technical artifacts

Line noise / Electromagnetic interference

As alternating current flows through the room’s electric wiring, it generates electromagnetic noise called “line noise”, that is picked up by the EEG cables with a frequency that depends on the local grid: 50 or 60 Hz depending on the country [10]. Modern amplifiers do a great job in reducing this noise, however it can still enter the recording.

Electronic equipment is also a source of line noise as an extension of the power grid. Electrodes closer to the source are more affected. Some devices can emit noise at varying frequencies depending on the state and properties of the device. For example, some dimmable lights generate interference at different frequencies depending on the light’s intensity.

→ Handling methods: Notch filters of 50 or 60 Hz. Other band-rejection filters if additional devices induce noise at different frequencies.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 8: Unfiltered data affected by line noise at 50 Hz. The spectral profile under the marked area is displayed for the highlighted channels. Diverse channels are affected to a different degree, but the noise is present in all channels.

Loose electrode contact

When the contact between the scalp and the electrode is disrupted, the signal becomes unstable. This can happen because of a loose-fitting cap, body movement or hair pushing the cap away. Loose contact more often leads to slow drifts in the signal. However, the electrochemical instabilities produced by the loose contact can lead to sudden conductance changes which manifest as an “electrode pop” in the data. This artifact can also affect ground and reference electrodes, which would consequently affect all other channels.

→ Handling methods: Artifact rejection for transient artifacts. Persistently unstable channels can be cautiously replaced through interpolation, or rejected entirely.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 9: Slow drifts in various channels with bad contact due to a loose-fitting cap. The drift depends on various conditions like the position of the electrode and how it contacts with the scalp.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 10: Electrode pop on Oz caused by movement during a mobile recording.

Cable movement

Movement of the EEG cables alters their conductive properties momentarily [12]. This produces transient signal alterations with varying shapes which depend on the type of cable movement. Most notably, cable swinging introduces oscillations at the frequency of the swing, which may overlap with EEG frequencies of interest. Modern EEG devices such as the actiCAP active electrode system, incorporate elements like amplification at the electrodes, which reduce cable movement artifacts [12].

→ Handling methods: Most cable-movement artifacts are discarded through artifact rejection. Swing artifacts may be attenuated through filtering. However, if the swinging overlaps with the EEG, it can hardly be separated from the data.

Artifacts in multi-modal recordings

The combination of EEG with other neuroscience techniques is now a frequent case in the field. However, many devices used in these other techniques unavoidably cause artifacts in the EEG.

In simultaneous EEG-fMRI, the scanner environment induces diverse artifacts. Most notably, the magnetic gradient switching induces large currents in the EEG leads, which produce large artifactual voltages [13]. Vibrations of the scanner further produce motion artifacts that are enhanced by the scanner’s magnetic field [14]. Artifact handling in EEG-fMRI requires special techniques that cope with the particular nature of these artifacts.

TMS pulses generate large spikes that may lead to the saturation of the EEG amplifier. Effects on the hardware furthermore lead to a decay artifact for a few milliseconds [15]. If there is amplifier saturation, it means that the EEG data is lost for a brief period, so the spike and decay are most frequently replaced by interpolated data.

Electrical stimulation like tDCS and tACS bring the stimulation current into the EEG. Removal of these artifacts is currently a work in progress in the research community.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 11: A. fMRI gradient artifact. B. TMS artifacts.

3. Artifact handling in BrainVision Analyzer 2

Analyzer 2 offers various tools for the two general artifact handling procedures: rejection and attenuation. A well-planned combination of the different techniques can largely improve signal quality to get the most information out of your EEG data.

3.1.   Artifact rejection

Raw Data Inspection and Artifact Rejection Transformation

During artifact rejection, those portions of data contaminated by artifacts are discarded from subsequent analyses. Beforehand, artifacts must be identified either by a human rater or a computer in an exercise of signal detection: ideally most artifacts are detected (hit) without letting artifacts pass (miss) and without mislabeling valid data (false alarm).

In Analyzer 2 this is accomplished through the Raw Data Inspection transformation for continuous data, or the Artifact Rejection transformation for segmented data. If the data is segmented, segments with artifacts can either be discarded or remain in the data with a “Bad Interval” marking, allowing for more flexibility in artifact handling. Both transformations offer a manual, automatic and a semi-automatic mode of inspection.

Manual Inspection allows the user to scroll through the data and mark artifacts. This mode offers great flexibility but the artifact demarcation is subjective and can be very time consuming. Manual inspection is most useful for marking conspicuous artifacts, or for short data sets.

Automatic Inspection lets the user define a set of objective criteria to search for artifacts automatically:

  • Gradient: detects steep changes in voltage within one millisecond (e.g. electrode pops).
  • Amplitude: looks for absolute voltages that surpass a threshold relative to zero (e.g. ocular artifacts, large muscular artifacts).
  • Max-Min: searches for relative changes in amplitude beyond a defined range. Useful to find artifacts within data that already contains an offset (e.g. within drifts, DC offset).
  • Low Activity: searches for flat lines by finding stretches of data with unnaturally little variation
Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 12: Artifact rejection criteria in Analyzer 2.

These criteria must be carefully calibrated to hit the maximum number of artifacts, with the least misses and false alarms possible. Automatic detection is fast and objective, but inflexible. Depending on the data, sometimes certain trade-offs must be accepted: restrictive criteria may find all artifacts but produce some false alarms; lenient criteria may not produce false alarms but will miss some artifacts.

Semi-Automatic Inspection lets the user verify the artifact detection achieved through automatic criteria, by adding or removing artifact markings to the selection. It then combines the objectivity of automatic inspection with the flexibility of the manual mode.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 13: Semi-Automatic artifact detection using the Raw Data Inspection transformation. Various artifacts including blinks, eye movements, muscular tension, drift and electrode pop are automatically detected. These can be confirmed, rejected or modified by the user.

Rejecting and interpolating channels

Sometimes a channel has an irreparably noisy signal throughout the entire recording. In this unfortunate situation, one option is to reject the channel entirely. In Analyzer this can be done with the Edit Channels transformation, where the channel in question can be disabled. An alternative approach is to replace the channel with an interpolated signal based on all other channels. This can be achieved through the Topographic Interpolation transformation. However, here the interpolated signal is only an estimation and should only be carefully interpreted, if at all.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 14: A channel which had loose contact is replaced by interpolated data.

3.2.   Artifact attenuation

During artifact attenuation, the data is processed to reduce the influence of artifacts on the EEG without discarding it. Ideally, artifacts should be reduced to negligible levels. Attenuation involves multiple methods, some of which are generic, and others devised for specific artifacts. Since these methods actively manipulate the data, calibrating them should be handled with care to prevent distorting or overcorrecting the EEG to the point of erasing the signal of interest.

Filtering. Analyzer’s IIR Filters transformation incorporates zero-phase-shift Butterworth filters to attenuate undesired frequencies.

  • High-pass filter: Attenuates frequencies below the low-cutoff. Common cutoffs in EEG are between 0.1 or 0.5 Hz to reduce drifts such as body sway, or skin potentials. Likewise, they dramatically reduce offsets as very slow oscillations in the data.
  • Low-pass filter: Reduces frequencies above the high-cutoff. As most EEG frequencies that are usually studied lie below 40 Hz, this is a common cutoff frequency. This reduces the impact of muscular artifacts and other unwanted high frequencies.
  • Notch filter: This is a pre-defined band rejection filter to attenuate line noise at 50 or 60 Hz.

In case of consistent noise at other frequencies, the Band Rejection transformation allows defining a custom range for a band rejection IIR filter.

Re-referencing. Offline re-referencing is a common preprocessing step that can also help with artifact handling. In Analyzer this is possible through the New Reference transformation. In particular, the Average Reference improves signal to noise ratio [16, 17], provided that sufficient channels are available to have good head coverage [18] and no bad channels are introduced into the average. This makes the Average Reference a common choice in mobile EEG studies [e.g. 19, 20].

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 15: Noise reduction after using the average reference (red) contrasted with before re-referencing (black). Notice how drift is suppressed and high frequency noise is attenuated at the edges while alpha oscillations are preserved.

Independent Component Analysis (ICA). EEG recordings include a mixture of signals from brain and non-brain sources. The ICA transformation in Analyzer [21] seeks to statistically separate this mixture into its components. With the Inverse ICA transformation, the components that represent artifacts can be discarded, to reconstruct the EEG without them. These are most commonly ocular artifacts, but can also be heartbeat, localized muscular tension or other artifacts that have a single source.

For optimal component separation, at least 64 channels are recommended. You must further be careful to adequately select the components to discard without removing valid EEG. Analyzer offers a user-friendly visual display to facilitate these decisions. Even after removing components, Inverse ICA can be reprocessed to change the component selection.

You can find more information about ICA in Analyzer in our ICA webinar. For a detailed explanation on how ICA works, please refer to this dedicated article.

Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 16: Interactive view during the inverse ICA transformation. ICA was used to remove body movement and ocular artifacts produced during a mobile EEG session. The data before correction (red) is overlayed on the corrected data (black) to monitor the effects of component removal before accepting the correction. To the right, the list of components and their topographic distribution is available. It is also possible to switch the view to check the time course of the components.

Methods for ocular artifact correction. Analyzer offers two methods of ocular correction that adapt to different conditions and requirements of the data.

  • The Ocular Correction transformation implements the method by Gratton & Coles [22]. The EOG channels are used to estimate the influence of blinks and eye movements on the EEG channels via regression. Ocular artifacts are then subtracted based on a correction factor. This method is fast to compute. However ideal performance requires recording EOG.
  • The Ocular Correction ICA transformation can be used with or without dedicated EOG channels. It offers various options to tailor ICA for eye movement correction. For instance, it includes blink detection, and sorts components in relation to ocular artifacts. As with regular ICA, at least 64 channels are recommended for optimal performance.
Getting to know EEG artifacts and how to handle them in BrainVision Analyzer 2

Figure 17: Attenuation of blink and eye movement artifacts using the Gratton & Coles (left) and the ICA-based (right) ocular correction methods. Corrections are represented by the overlayed red traces. The EOG channels are included underneath to better locate the ocular artifacts.

Specialized artifact correction tools. Some artifacts are so peculiar that they require removal techniques specific for them. This is particularly the case in the realm of simultaneous EEG-fMRI, for which Analyzer offers tailored transformations to handle gradient, cardio-ballistic and motion artifacts. These transformations are reviewed in detail in our webinar channel.

4. Planning strategies for artifact handling

With multiple artifact handling tools at your disposal, a well-planned strategy is the best complement to combine them to get the best results. Part of this strategy starts already during experimental design. As an example, let us imagine that we want to assess the visually evoked potential in a neurological patient group. Under this condition, we would expect a greater number of motion and ocular artifacts in the data. For handling motion artifacts, we plan to have a larger number of trials to ensure good ERP quality. However, since we do not want to tire the participants with a long session, we opt for using 32 electrodes rather than 64 to speed up the preparation.

The question that follows is how to handle the expected ocular artifacts. Given the expected increase in movement artifacts and that we already opted for 32 electrodes, using ICA would entail the risk of not achieving optimal component separation. We can then resort to ocular correction by the Gratton & Coles method. As this method works best with bipolar EOG, we decide to include it into our EEG measurement. Later during data processing, before ocular correction we first perform filtering and reject large non-ocular artifacts from the EOG channels, such as muscle twitches or electrode movement. This will provide a cleaner EOG signal for input, improving ocular correction and preventing data distortion.

This is one of many possible strategies in countless possible scenarios. You can find further examples in our artifact handling webinar. There is no single strategy that fits all scenarios, and the processing pipeline needs to be customized for each study. A good starting point for devising your own strategy is previous literature from the same field. As in the example above, it is also helpful to know the properties and requirements of the planned artifact handling techniques. Testing the strategy on pilot data is the best way to evaluate its effects before deciding for a final pipeline.

Concluding remarks

BrainVision Analyzer 2 offers a full set of EEG artifact rejection and attenuation tools for handling a wide variety of artifacts. You can combine these tools to great effect within an integral artifact handling strategy that often starts from experimental design. If you have questions about artifact handling in your data, you can always contact us at support@brainproducts.com.

References

1.       Croft, R. J., & Barry, R. J. (2000). Removal of ocular artifact from the EEG: a review. Clinical Neurophysiology, 30(1), 5-19. https://doi.org/10.1016/S0987-7053(00)00055-1
2.       Lins, O. G., Picton, T. W., Berg, P., & Scherg, M. (1993). Ocular artifacts in EEG and event-related potentials I: Scalp topography. Brain Topography, 6(1), 51-63. https://doi.org/10.1007/BF01234127
3.       Luck, S. J. (2014). An introduction to the event-related potential technique. MIT press.
4.       Kobler, R. J., Sburlea, A. I., Lopes-Dias, C., Schwarz, A., Hirata, M., & Müller-Putz, G. R. (2020). Corneo-retinal-dipole and eyelid-related eye artifacts can be corrected offline and online in electroencephalographic and magnetoencephalographic signals. Neuroimage, 218, 117000. https://doi.org/10.1016/j.neuroimage.2020.117000
5.     Muthukumaraswamy, S. D. (2013). High-frequency brain activity and muscle artifacts in MEG/EEG: a review and recommendations [Review]. Frontiers in Human Neuroscience, 7, 138. https://doi.org/10.3389/fnhum.2013.00138
6.     Kappel, S. L., Looney, D., Mandic, D. P., & Kidmose, P. (2017). Physiological artifacts in scalp EEG and ear-EEG. BioMedical Engineering OnLine, 16(1), 103. https://doi.org/10.1186/s12938-017-0391-2
7.       Chen, X., Xu, X., Liu, A., Lee, S., Chen, X., Zhang, X., McKeown, M. J., & Wang, Z. J. (2019). Removal of Muscle Artifacts From the EEG: A Review and Recommendations. IEEE Sensors Journal, 19(14), 5353-5368. https://doi.org/10.1109/jsen.2019.2906572
8.       Mullinger, K. J., Havenhand, J., & Bowtell, R. (2013). Identifying the sources of the pulse artefact in EEG recordings made inside an MR scanner. Neuroimage, 71, 75-83. https://doi.org/10.1016/j.neuroimage.2012.12.070
9.       De Vos, M., Deburchgraeve, W., Cherian, P. J., Matic, V., Swarte, R. M., Govaert, P., Visser, G. H., & Van Huffel, S. (2011). Automated artifact removal as preprocessing refines neonatal seizure detection. Clinical Neurophysiology, 122(12), 2345-2354. https://doi.org/10.1016/j.clinph.2011.04.026
10.    Kappenman, E. S., & Luck, S. J. (2010). The effects of electrode impedance on data quality and statistical significance in ERP recordings. Psychophysiology, 47(5), 888-904. https://doi.org/10.1111/j.1469-8986.2010.01009.x
11.    Protzak, J., & Gramann, K. (2018). Investigating Established EEG Parameter During Real-World Driving [Original Research]. Frontiers in Psychology, 9(2289). https://doi.org/10.3389/fpsyg.2018.02289
12.    Symeonidou, E. R., Nordin, A. D., Hairston, W. D., & Ferris, D. P. (2018). Effects of Cable Sway, Electrode Surface Area, and Electrode Mass on Electroencephalography Signal Quality during Motion. Sensors (Basel), 18(4). https://doi.org/10.3390/s18041073
13.    Chowdhury, M. E. H., Khandakar, A., Mullinger, K. J., Al-Emadi, N., & Bowtell, R. (2019). Simultaneous EEG-fMRI: Evaluating the Effect of the EEG Cap-Cabling Configuration on the Gradient Artifact. Frontiers in Neuroscience 13, 690. https://doi.org/10.3389/fnins.2019.00690
14.    Bullock, M., Jackson, G. D., & Abbott, D. F. (2021). Artifact Reduction in Simultaneous EEG-fMRI: A Systematic Review of Methods and Contemporary Usage [Systematic Review]. Frontiers in Neurology, 12, 622719. https://doi.org/10.3389/fneur.2021.622719
15.    Litvak, V., Komssi, S., Scherg, M., Hoechstetter, K., Classen, J., Zaaroor, M., Pratt, H., & Kahkonen, S. (2007). Artifact correction and source analysis of early electroencephalographic responses evoked by transcranial magnetic stimulation over primary motor cortex [Research Support, Non-U.S. Gov’t]. Neuroimage, 37(1), 56-70. https://doi.org/10.1016/j.neuroimage.2007.05.015
16.    Tsuchimoto, S., Shibusawa, S., Iwama, S., Hayashi, M., Okuyama, K., Mizuguchi, N., Kato, K., & Ushiba, J. (2021). Use of common average reference and large-Laplacian spatial-filters enhances EEG signal-to-noise ratios in intrinsic sensorimotor activity. Journal of Neuroscience Methods, 353, 109089. https://doi.org/10.1016/j.jneumeth.2021.109089
17.    Bigdely-Shamlo, N., Mullen, T., Kothe, C., Su, K.-M., & Robbins, K. A. (2015). The PREP pipeline: standardized preprocessing for large-scale EEG analysis [Methods]. Front Neuroinform, 9, 16. https://doi.org/10.3389/fninf.2015.00016
18.   Hu, S., Lai, Y., Valdes-Sosa, P. A., Bringas-Vega, M. L., & Yao, D. (2018). How do reference montage and electrodes setup affect the measured scalp EEG potentials? Journal of Neural Engineering, 15(2), 8406–8420. https://doi.org/10.1088/1741-2552/aaa13f
19.    Klug, M., & Gramann, K. (2021). Identifying key factors for improving ICA-based decomposition of EEG data in mobile and stationary experiments. European Journal of Neuroscience, 54(12). https://doi.org/10.1101/2020.06.02.129213
20.    Arad, E., Bartsch, R. P., Kantelhardt, J. W., & Plotnik, M. (2018). Performance-based approach for movement artifact removal from electroencephalographic data recorded during locomotion. PLoS One, 13(5), e0197153. https://doi.org/10.1371/journal.pone.0197153
21.   Bell, A. J., & Sejnowski, T. J. (1995). An information-maximization approach to blind separation and blind deconvolution. Neural Comput, 7(6), 1129-1159. https://doi.org/10.1162/neco.1995.7.6.1129
22.    Gratton, G., Coles, M. G. H., & Donchin, E. (1983). A new method for off-line removal of ocular artifact. Electroencephalogrography and Clinical Neurophysiology, 55(4), 468-484. https://doi.org/10.1016/0013-4694(83)90135-9