All about EEG artifacts and filtering tools

One of the main challenges in working with electroencephalographic (EEG) data is ensuring that the recorded signals are clean and exhibit a high signal-to-noise ratio (SNR). Because EEG amplitudes are typically in the microvolt range, they are highly susceptible to various sources of contamination, commonly referred to as artifacts.

These unwanted signals can obscure the underlying neural activity and compromise the quality of the data, making artifact detection and removal essential for accurate analysis and reliable applications. In this article, we examine the most common types of EEG artifacts and the primary techniques and tools employed to minimize their impact and recover meaningful brain signals.

What is an EEG artifact?

Neurons communicate by generating electrical signals, but a single neuron’s activity is too weak to measure from the scalp. Only when millions of neurons fire synchronously does the resulting electric field become strong enough to pass through tissue, bone, and hair and reach EEG electrodes, though it remains significantly attenuated (Sörnmo & Laguna, 2005; Nunez & Srinivasan, 2006). 

An EEG artifact is any recorded signal that does not originate from neural activity. This includes physiological sources (such as eye movements, muscle activity, and heartbeats) and non-physiological sources including electrical interference, electrode issues, and movement artifacts (Urigüen & Garcia-Zapirain, 2015). Artifacts contaminate EEG recordings by injecting brain signals, which can mimic or obscure genuine neural activity and significantly reduce signal quality (Xiong, 2024).

“Artifacts are signals recorded by EEG but not generated by brain. Some artefacts may mimic true epileptiform abnormalities or seizures. Awareness of the logical topographic field of distribution for true EEG abnormality is important in distinguishing artifact from brain waves. Physiologic artifacts originate from the patient, and non-physiologic artifacts originate from the environment of the patient.” (EEG Artifacts, Springer)

Quick FAQ: EEG Artifacts and Signal Cleaning

Q: What makes EEG data difficult to work with?

EEG signals are measured in microvolts and are extremely sensitive to contamination. This means they can easily be affected by internal body processes or external interference, which introduces noise known as artifacts.

Q: What is an EEG artifact?

An artifact is any component of the EEG signal that does not originate from the brain. These can be physiological (e.g., blinking, muscle activity) or non-physiological (e.g., cable movement, AC interference).

Q: Why is artifact removal important?

Artifacts can distort or mask genuine neural signals. This not only reduces data quality but can lead to misinterpretation or even clinical misdiagnosis.

Types of EEG artifacts 

Identifying artifact types is the first step toward effectively removing them. EEG artifacts are generally categorized by their origin: 

Physiological artifacts (originated from the body):

• Ocular activity:  Eye blinks or movement distort scalp recordings through corneo-retinal potentials and eyelid effects.
• Muscle activity: Facial or neck muscle contractions produce broadband noise overlapping EEG frequencies (20–300 Hz).
• Cardiac activity and ballistocardiogram (BCG): Pulsatile signals from the heart or within EEG–fMRI setups can introduce rhythmic artifacts.
• Respiration and Perspiration: Slow drifts or electrode impedance changes due to breathing or sweat can contaminate EEG.

Non-physiological (technical) artifacts:

• Electrode pop and cable movement: Sudden impedance changes cause transient spikes.
• Incorrect reference placement or poor contact: Leads to baseline shifts or exaggerated noise.
• AC power and electromagnetic interference: Ambient 50/60 Hz coupling common in non-shielded environments.
• Subject motion: Head or body movements can introduce large, non-linear noise bursts.

Modern EEG recordings, such as with Bitbrain’s 16 channel system filtered at 0.5–30 Hz, often rely on Independent Component Analysis (ICA) or deep learning CNN–LSTM models to detect and isolate these artifacts at scale. Accurate identification and removal of these artifacts are crucial (Amin, 2023; Zhang, 2025). Failures in this process can bias data interpretation, obscure neural signals, and even lead to clinical misdiagnosis (e.g., confusing artifacts with epileptiform activity or sleep rhythms).

Physiological artifacts

Physiological artifacts originate from the body’s own activity and are not related to brain signals, but they frequently contaminate EEG recordings. Below are the most common sources, their origin, impact, and signature in the time and frequency domains.

Ocular activity (EOG Artifact)

• Origin: The eye behaves like an electric dipole due to the charge difference between the cornea and retina. When the eye moves, this dipole shifts, generating a field disturbance measurable on the scalp.
• Why it affects EEG:  This disturbance, called the Electrooculogram (EOG), typically reaches 100–200 µV, often an order of magnitude larger than EEG signals.
• Typical causes: Eye blinks, saccades, lateral gaze movements.
  
• Time-domain effect: Sharp, high-amplitude deflections, especially over frontal electrodes (e.g., Fp1, Fp2). Lateral movements affect electrodes near the temples. Artifact amplitude is proportional to the gaze angle.
  
• Frequency-domain effect: Dominant in low frequencies, especially in the delta (0.5–4 Hz) and theta (4–8 Hz) bands, potentially mimicking cognitive processes.

Muscle activity (EMG Artifact)

• Origin: Muscle contractions generate electric signals, recorded as Electromyography (EMG).
• Why it affects EEG: MG signals are broadband and high in frequency, often overlapping EEG rhythms and introducing significant noise.
• Typical causes: Clenching jaw, neck tension, swallowing, chewing, talking, frowning, sniffing, or even hiccupping.
• Time-domain effect: High-frequency noise superimposed on EEG, with amplitude proportional to contraction strength.
• Frequency-domain effect: Artifacts dominate beta (13–30 Hz) and gamma (>30 Hz) ranges, masking important cognitive and motor activity signals.
  

Cardiac activity (ECG or Pulse Artifact)

• Origin: The heart’s electrical signal, or Electrocardiogram (ECG), sometimes appears on scalp EEG.
• Why it affects EEG: Though usually weak, ECG artifacts become visible depending on body shape, electrode location, or amplifier sensitivity.
• Typical causes: Heartbeats (pulse artifact).
• Time-domain effect: Rhythmic waveforms recurring at the heart rate, often in central or neck-adjacent channels.
• Frequency-domain effect: ECG overlaps several EEG bands and may go undetected without cross-channel analysis or an ECG reference.

Perspiration (Sweat Artifact)

Origin: Activity from sweat glands modifies local electrode impedance and creates potential shifts.
Why it affects EEG: Perspiration introduces slow baseline drifts or shorts between electrodes, especially during physical activity or in high temperatures.
Typical causes: Include heat, stress, and long-duration recordings.
Time-domain effect: Slow potential shifts, apparent over long epochs.
Frequency-domain effect: Contaminates the delta and theta bands, impairing sleep and low-frequency cognitive assessments.

Respiration

• Origin: Movements of the chest and head during breathing, particularly when lying down.
• Why it affects EEG: Breathing alters electrode-skin contact, especially in sleep studies.
• Typical causes: Include deep breathing and sleep respiration cycles.
• Time-domain effect: Slow waveforms synchronized with respiration rate (e.g., 12–20 breaths per minute).
• Frequency-domain effect: Mainly affects low-frequency bands, overlapping delta and theta rhythms.

Non-Physiological (Technical) EEG Artifacts

Unlike physiological artifacts, technical artifacts originate from external or mechanical sources, such as hardware malfunctions, environmental interference, or improper setup. These can introduce significant distortions that are often mistaken for neural activity.

Electrode Pop

• Origin: Caused by a temporary disruption in contact between an electrode and the scalp. This can result from physical contact with the sensor, cable motion, or spontaneous changes in skin-electrode impedance, often due to drying gel or sweat accumulation.
• Why it affects EEG: Sudden shifts in contact potential lead to transient voltage spikes, unrelated to brain activity.
• Typical causes: Include touching the cap, head movement, pulling the electrode cable, or drying the electrolyte gel.
• Time-domain effect: Appears as abrupt, high-amplitude transients, often isolated to a single channel. The morphology can vary, ranging from sharp spikes to complex waveform distortions.
• Frequency-domain effect: Difficult to characterize consistently, as electrode pops produce broadband, non-stationary noise. Their irregularity challenges both manual and automated detection methods.

Cable Movement

• Origin: Occurs when electrode or amplifier system cables shift during a recording, whether due to participant movement, loose connectors, or external contact with the wiring.
• Why it affects EEG: Movement can cause electromagnetic interference and alter the impedance of scalp electrodes, especially in high-impedance systems.
• Typical causes: Tugging or brushing cables, participant repositioning, or external contact with electrode wires.
• Time-domain effect: Highly variable. If movement is rhythmic, the resulting artifacts can produce repetitive waveforms that mimic neural oscillations or eye blinks. Non-rhythmic cable shifts often create sudden deflections or drift.
• Frequency-domain effect: Depending on the periodicity of movement, cable artifacts can introduce artificial spectral peaks at low or mid frequencies, potentially mistaken for delta or alpha rhythms.

Incorrect Reference Placement

• Origin: Occurs when the reference electrode is not placed, improperly connected, or suffers from poor contact with the scalp.
• Why it affects EEG: Since all EEG channels are measured relative to the reference, a faulty reference results in signals that do not reflect actual brain activity, but instead amplify noise or baseline drift across all channels.
• Typical causes: Omitted reference electrode, dried conductive gel, loose connections, or excessive impedance at the reference site.
• Time-domain effect: Results in abrupt, high-amplitude shifts across all channels. When the reference is corrected mid-recording, signals typically return to normal values gradually due to the filtering and stabilization process.
• Frequency-domain effect: Produces abnormally high power across all channels, often accompanied by non-physiological peaks unrelated to actual EEG activity.

AC Electrical and Electromagnetic Interference

• Origin: Generated by nearby alternating current (AC) power lines, electrical devices, or wiring that emit electromagnetic fields.
• Why it affects EEG: When electrode cables lack proper shielding or when recordings are performed in electrically noisy environments, ambient electromagnetic radiation can couple into the EEG system, contaminating the signal.
• Typical causes: Fluorescent lights, monitors, power adapters, unshielded EEG cables, or proximity to AC mains wiring.
• Time-domain effect: Appears as persistent high-frequency noise that overlays the EEG signal, sometimes modulating with nearby device activity.
• Frequency-domain effect: Produces a sharp, narrow spectral peak at 50 Hz or 60 Hz, depending on the local power grid standard (e.g., 50 Hz in Europe, 60 Hz in North America). This is commonly referred to as line noise or a power-line artifact.

Body Movements

• Origin: Caused by gross motor activity, primarily head movements, but also includes arm swings, walking, or postural shifts.
• Why it affects EEG: Even subtle movements can disrupt the electrode–skin interface, altering contact impedance and distorting the EEG signal. In mobile or ambulatory EEG systems, movement artifacts are prevalent and challenging to distinguish from other signals.
• Typical causes: Head tilts, nodding, arm gestures, walking, or running during recordings.
• Time-domain effect: Manifests as slow, transient waveforms that often correlate with the rhythm of physical movement. These artifacts may resemble neural slow waves if not correctly identified.
• Frequency-domain effect: Primarily affects low-frequency bands, overlapping with delta (0.5–4 Hz) and theta (4–8 Hz) activity, which can obscure signals related to sleep, fatigue, or cognitive processing.

EEG Artifact Filtering Techniques (Data-Driven Approaches)

When processing EEG data, managing artifacts effectively is essential for extracting meaningful neural signals. Depending on the analysis goals, four major strategies are commonly used:

1. EEG artifact Rejection

This method involves identifying and excluding entire EEG segments (epochs) that contain artifacts. Traditionally, patterns are detected manually by experts or flagged by automated algorithms based on statistical deviations in the time or frequency domain. A recent approach, LSTEEG, introduces a deep learning-based autoencoder architecture using LSTMs to detect and correct EEG artifacts with high precision (Aquilué-Llorens & Soria-Frisch, 2025). It has been shown to outperform convolutional autoencoders in both artifact detection and correction tasks.

• Strength: Highly effective at removing contamination.
• Limitation: Discards both artifact and valuable EEG data, making it inefficient, especially for short or sparse recordings where data preservation is crucial.

2. Filtering

Filtering techniques aim to remove specific artifact frequencies or signal features while preserving the underlying EEG. These include:

• Linear band-stop filters to eliminate power-line noise (e.g., 50/60 Hz artifacts).
• Regression-based filters that subtract known artifact sources, such as EOG or ECG, using reference channels.
• Adaptive filtering with a reference signal, such as eye movement or heartbeat.
• Wiener filters and Bayesian filters for dynamic denoising.

These methods assume that the EEG signal is a mixture of neural activity and structured noise. For example, using an EOG channel, a regression model estimates the artifact's influence on each EEG channel and subtracts it accordingly.

• Strength: Preserves a large portion of EEG content.
• Limitation: May underperform if the artifact overlaps heavily with EEG bands of interest or if reference signals are missing.

3. Blind Source Separation (BSS)

Blind Source Separation methods decompose multichannel EEG into independent or orthogonal sources, assuming the recorded signal is a linear mix of underlying components. The most prevalent method is Independent Component Analysis (ICA) (Choi et al., 2005), which separates EEG into independent components, allowing artifacts such as EOG or EMG to be isolated from actual brain activity. 

Recent findings, however, indicate that ICA's benefits may be modest for deep learning–based decoding tasks, and its utility may depend on dataset size, channel count, and classifier type (Kang et al., 2024). In contrast, newer algorithms focus on real-time artifact correction:

A fast, ongoing BSS algorithm successfully removed ~88% of ocular, cardiac, muscle, and line-noise artifacts in continuous EEG, showing promise for real-time BCI applications (Ille et al., 2024). 

The Artifact Removal Transformer (ART), a recent transformer-based deep model, demonstrated superior performance over state-of-the-art methods—including ICA—across multiple artifact types (Chuang et al., 2025).

Artifact-laden components can be discarded manually or classified automatically using tools such as MARA, iMARA (for infant EEG, with>75% expert agreement), ICLabel, or deep-learning classifiers, thereby enabling cleaner EEG reconstruction.

• Pros: Eliminates multiple artifact types without auxiliary channels; well-suited for high-density EEG; emergent real-time BSS techniques enhance usability.
• Cons: Performance drops with limited channel count or short sessions; potential removal of neural signals if artifact classification is incorrect.

The advantage of Blind Source Separation is that it operates without requiring reference channels or prior knowledge about the nature of the artifacts. This makes it highly versatile, especially for removing multiple types of contamination in a single step.

However, its main limitation is that BSS techniques like ICA rely on global multichannel decomposition—they analyze the full EEG data matrix rather than filtering on a per-channel basis. As a result, their performance tends to degrade when using low-density EEG systems or when the dataset is short or incomplete, making the separation of independent sources less reliable.

4. Source Decomposition Methods

These techniques decompose each EEG channel into basic waveforms, allowing for the selective removal of components contaminated by artifacts before reconstructing the clean signal. The most established approach is Wavelet Decomposition (Issa, 2019), while newer but less widely adopted variants include Empirical Mode Decomposition (EMD) (Rakhmatulin, 2024) and Nonlinear Mode Decomposition (NMD) (Iatsenko et al., 2015).

In wavelet decomposition, the EEG signal is transformed into coefficients representing various frequency scales and drifts, based on a chosen “mother” wavelet. Artifact filtering is done by thresholding these coefficients and then reconstructing the signal.

• Advantages: Enables artifact removal at the channel level, preserving spatial specificity.
• Limitations: Requires the accurate selection of base functions (e.g., wavelets or modes), and improper decomposition can result in the loss of genuine EEG data. These methods are also more computationally intensive and still under active development.

EEG Artifact Filtering Techniques: Online vs. Offline Approaches

An essential consideration in EEG artifact filtering is whether the technique operates offline or online (Ille, 2024; Longo, 2025).

Offline methods typically require manual intervention and expert supervision, making them unsuitable for real-time or autonomous applications. Common examples include visual inspection to reject noisy EEG segments or manual selection of artifact-related components after decomposition. These approaches, while accurate, are time-consuming and non-scalable.

In contrast, online methods are automated and can be integrated into real-time systems. Techniques such as regression-based filtering or adaptive filters using reference signals (e.g., EOG, ECG) are well-suited for online operation. More advanced methods like Blind Source Separation (e.g., ICA) and Wavelet or Mode Decomposition can also run online when combined with automated artifact detection based on statistical thresholds or pre-learned templates from clean EEG data.
By enabling real-time operation, online filtering is crucial for brain-computer interfaces (BCIs), neurofeedback, and other applications that require continuous signal processing.

EEG Artifact Software

There are several toolboxes and libraries available for EEG signal filtering. Here we are going to focus on a small subset of them that are probably the most used ones at the time this document was written. All of them are software libraries that can be used independently of the EEG system that acquires the data:

1. EEGLAB (EEGLAB, EEGLAB Wiki): This is an interactive Matlab toolbox for processing continuous and event-related EEG, MEG and other electrophysiological data. It includes filtering techniques such as independent component analysis (ICA) or artifact rejection and several filtering plugins can be downloaded to increase the toolbox potential. It also incorporates time/frequency analysis, event-related statistics, and several modes of visualization of the averaged and single-trial data. EEGLAB runs on Windows, Mac OS X, Linux and Unix. 
2. FieldTrip (FieldTrip toolbox): This is a MATLAB toolbox for MEG, EEG, iEEG and NIRS analysis. It offers preprocessing techniques and analysis methods, such as time-frequency analysis or source reconstruction using dipoles. It supports the data formats of all major MEG systems and of the most popular EEG, iEEG and NIRS systems and new data formats can be added easily. You can implement your own analysis protocols in a MATLAB script using FieldTrip high-level functions. FieldTrip  is an open source software under the GNU general public license.
3. MNE (MNE — MNE 0.20.0 documentation): Open-source Python software for exploring, visualizing, and analyzing human neurophysiological data: MEG, EEG, sEEG, ECoG, and more. The software has a growing community behind and several python packages has been developed to add a graphical user interface, automatic bad channel detection and interpolation, independent component analysis (ICA), connectivity analysis, general-purpose statistical analysis of MEG/EEG signals or a python implementation of the Preprocessing Pipeline (PREP) for EEG data among others.

Other available alternatives include: 

• AnEEG: A deep learning-based artifact removal tool using an LSTM-driven GAN architecture. It significantly outperforms traditional methods and processes multi-channel EEG in real-time environments.
• EEG-Denoise / DTP-Net: An open-source TensorFlow implementation of a multi-scale CNN-autoencoder designed for single-channel EEG artifact removal in the time–frequency domain. It reconstructs clean signals with high Fidelity.
• Artifact Removal Transformer (ART): A transformer-based end-to-end model that outperforms conventional methods—including ICA—across multiple artifact types and datasets.
• Edge-ready Autoencoder (e.g., Arduino, Coral): An efficient deep autoencoder capable of removing artifacts “on the edge” (embedded hardware) for real-time, wireless EEG systems.
• BEST Toolbox: A Python package from the Mayo Clinic for long-term invasive and non-invasive EEG processing. Includes modules for artifact removal (e.g., DBS, movement) alongside sleep staging and feature extraction.
• SleepEEGpy: A Python framework for preprocessing and artifact detection specifically in sleep EEG, supporting integrated workflows from raw data to cleaned, sleep-ready signals.
• Chronux: MATLAB/C package focused on advanced time-series processing—supports multitaper spectral analysis with tools to identify and mitigate artifacts.
• NeuroKit2: A Python library for physiological signal processing, including EOG, ECG, EMG artifact detection and removal, widely used across modalities.

Which Should You Choose?

Recommended software for each use case:

End-to-end denoising: AnEEG, ART, DTP-Net
Edge computing: Edge-ready autoencoders (Arduino, Coral)
Long-term monitoring: BEST Toolbox, SleepEEGpy
Spectral & statistical analysis: Chronux, NeuroKit2

Conclusions

Dealing with EEG artifacts is not just a preprocessing step, it’s a critical determinant of data quality, reliability, and downstream interpretation. From physiological sources, such as eye blinks and muscle tension, to technical interferences like poor referencing or AC noise, artifacts can significantly distort brain signals and compromise both research findings and clinical decisions.

Fortunately, the field has moved far beyond manual rejection. Today’s EEG workflows increasingly rely on advanced, automated solutions—from adaptive filters and regression models to ICA, wavelet decomposition, and cutting-edge deep learning tools like LSTM autoencoders and transformer-based denoisers. These technologies not only improve artifact detection and removal but also enable real-time EEG cleaning in portable and embedded systems.

The key takeaway? Effective artifact management requires a tailored strategy that balances the preservation of neural signals with the elimination of unwanted noise. Whether you're designing a BCI, conducting sleep studies, or analyzing cognitive states, selecting the right combination of techniques and tools is crucial to unlocking the full potential of EEG data.

You might also be interested in:

• How Deep Learning is changing machine learning AI in EEG data processing 
• EEG Electrode Placement Options 
• EEG test: Uses, Procedures and Risks 
• Main features of EEG amplifiers 
• What is EEG and what is it used for?  
• How to Select a Dry-EEG Headset for your Research Application 
• Redefining brain activity monitoring with pioneering EEG textile technology 
• EEG and Multimodal Research in Neuroscience 
  
• Exploring the Science of Mind Wandering with Dr. Julia Kam 
• A Window into the Social Brain: How Collective Rituals Shape Neural Connection 
• The Use of EEG for ADHD Diagnosis and Treatment
• The Ultimate Guide of Technical Features of EEG Systems
• EEG Synchronization With Other Biosensors and Software
• Connectivity Layer and Other Features of EEG Headsets Explained
• The Wet-EEG Cap & Differences Between Semi-Dry, Saline or Gel EEG caps
• An introduction to brain-computer interface using EEG signals
• Advances in motor neuroprosthetics improve mobility in tetraplegics
• Bringing BCIs to the user’s home for neurorehabilitation and assistive applications

References

• Amin U, Nascimento FA, Karakis I, Schomer D, Benbadis SR. (2023) Normal variants and artifacts: Importance in EEG interpretation. Epileptic Disord. Oct;25(5):591-648. doi: 10.1002/epd2.20040.
• Aquilué-Llorens, D., & Soria-Frisch, A. (2025). EEG artifact detection and correction with deep autoencoders. arXiv preprint. https://doi.org/10.48550/arXiv.2502.08686
• Choi, S., Cichocki, A., Park, H. M., & Lee, S. Y. (2005). Blind source separation and independent component analysis: A review. Neural Information Processing—Letters and Reviews, 6(1), 1–57.
• Chuang, C. H., Chang, K. Y., Huang, C. S., & Bessas, A. M. (2025). Augmenting brain-computer interfaces with ART: An artifact removal transformer for reconstructing multichannel EEG signals. NeuroImage, 310, 121123. https://doi.org/10.1016/j.neuroimage.2025.121123
• Iatsenko, D., McClintock, P. V. E., & Stefanovska, A. (2015). Nonlinear mode decomposition: A noise-robust, adaptive decomposition method. Physical Review E, 92(3), 032916.https://doi.org/10.1103/PhysRevE.92.032916
• Ille, N., Nakao, Y., Yano, S., Taura, T., Ebert, A., Bornfleth, H., Asagi, S., Kozawa, K., Itabashi, I., Sato, T., Sakuraba, R., Tsuda, R., Kakisaka, Y., Jin, K., & Nakasato, N. (2024). Ongoing EEG artifact correction using blind source separation. Clinical Neurophysiology, 158, 149 158. https://doi.org/10.1016/j.clinph.2023.12.133
• Issa, M. F., & Juhasz, Z. (2019). Improved EOG artifact removal using wavelet enhanced independent component analysis. Brain Sciences, 9(12), 355. https://doi.org/10.3390/brainsci9120355
• Kang, T., Chen, Y., & Wallraven, C. (2024). I see artifacts: ICA-based EEG artifact removal does not improve deep network decoding across three BCI tasks. Journal of Neural Engineering, 21(6). https://doi.org/10.1088/1741-2552/ad788e
• Longo, L., & Reilly, R. B. (2025). onEEGwaveLAD: A fully automated online EEG wavelet-based learning adaptive denoiser for artefacts identification and mitigation. PLoS ONE, 20(1), e0313076. https://doi.org/10.1371/journal.pone.0313076
• Miran, S., Akram, S., Sheikhattar, A., Simon, J. Z., Zhang, T., & Babadi, B. (2018). Real-time tracking of selective auditory attention from M/EEG: A Bayesian filtering approach. Frontiers in Neuroscience, 12, 262. https://doi.org/10.3389/fnins.2018.00262
• Nunez, P. L., & Srinivasan, R. (2006). Electric fields of the brain: The neurophysics of EEG (2nd ed.). Oxford University Press. ISBN 9780199865673
• Panych, L. P., Wada, J. A., & Beddoes, M. P. (1989). Practical digital filters for reducing EMG artefact in EEG seizure recordings. Electroencephalography and Clinical Neurophysiology, 72(4), 268–276.
• Rakhmatulin, I. (2024). Encoder with the empirical mode decomposition (EMD) to remove muscle artefacts from EEG signal. arXiv preprint. https://arxiv.org/pdf/2409.14571
• Ronca, V., Capotorto, R., Di Flumeri, G., Giorgi, A., Vozzi, A., Germano, D., Virgilio, D. V., Borghini, G., Cartocci, G., Rossi, D., Inguscio, B. M. S., Babiloni, F., & Aricò, P. (2024). Optimizing EEG signal integrity: A comprehensive guide to ocular artifact correction. Bioengineering, 11(10), 1018. https://doi.org/10.3390/bioengineering11101018
• Rosanne, O., Albuquerque, I., Cassani, R., Gagnon, J. F., Tremblay, S., & Falk, T. H. (2021). Adaptive filtering for improved EEG-based mental workload assessment of ambulant users. Frontiers in Neuroscience, 15, 611962. https://doi.org/10.3389/fnins.2021.611962
• Sazgar, M., & Young, M. G. (2019). EEG artifacts. In Absolute epilepsy and EEG rotation review (pp. 107–114). Springer. https://doi.org/10.1007/978-3-030-03511-2_8
• Sörnmo, L., & Laguna, P. (2005). Bioelectrical signal processing in cardiac and neurological applications. Elsevier Science & Technology. ISBN 9780124375529
• Urigüen, J. A., & Garcia-Zapirain, B. (2015). EEG artifact removal—State-of-the-art and guidelines. Journal of Neural Engineering, 12(3), 031001. https://doi.org/10.1088/1741-2560/12/3/031001
• Węsierski, D., Rufuie, M. R., Milczarek, O., Ziembla, W., Ogniewski, P., Kołodziejak, A., & Niedbalski, P. (2023). Rating by detection: An artifact detection protocol for rating EEG quality with average event duration. Journal of Neural Engineering, 20(2). https://doi.org/10.1088/1741-2552/acbabe
• Xiong, W., Ma, L., & Li, H. (2024). A general dual-pathway network for EEG denoising. Frontiers in Neuroscience, 17, 1258024. https://doi.org/10.3389/fnins.2023.1258024
• Zhang, G., & Luck, S. J. (2025). Assessing the impact of artifact correction and artifact rejection on the performance of SVM- and LDA-based decoding of EEG signals. NeuroImage, 316, 121304. https://doi.org/10.1016/j.neuroimage.2024.1