PIPEMAT-RS: Development and Validation of a Standardized MATLAB Pipeline for Resting-State EEG Preprocessing

Summary

This protocol presents PIPEMAT-RS, a standardized MATLAB-based preprocessing pipeline for resting-state EEG data. It ensures artifact removal, improves signal quality, and enhances data reproducibility across studies. The pipeline automates key preprocessing steps, including filtering, independent component analysis (ICA), and artifact classification, facilitating consistent and reliable EEG analysis for neurophysiological research.

Abstract

Electroencephalography (EEG) is a crucial tool in neuroscience research and clinical applications, but raw EEG data often contain noise and artifacts that compromise signal quality. To address this, we developed PIPEMAT-RS, a standardized MATLAB-based preprocessing pipeline for resting-state EEG data. PIPEMAT-RS follows a structured seven-step workflow: file format conversion, EEG montage configuration, downsampling, filtering, artifact rejection, independent component analysis (ICA), and ICLabel classification for automated artifact removal. This protocol enhances EEG data quality by minimizing human intervention while maintaining high accuracy in artifact rejection. It was validated using multiple datasets, demonstrating its robustness in improving signal integrity. PIPEMAT-RS provides a systematic approach that facilitates reproducibility and reliability in EEG studies, aligning with commonly adopted practices in the field and offering a clearly documented structure that can complement existing pipelines. By standardizing EEG preprocessing, PIPEMAT-RS facilitates neurophysiological research and clinical applications, allowing for more accurate interpretations of resting-state brain activity and its associations with neurological and psychiatric conditions.

Introduction

The electroencephalogram (EEG) is a critical tool in both neuroscience research and clinical practice, offering valuable insights into brain electrical activity. Resting-state EEG data, in particular, are extensively utilized to investigate various neurological and psychiatric conditions, such as stroke, fibromyalgia, and chronic neuropathic pain. Despite its widespread application, the analysis of EEG data necessitates rigorous preprocessing to eliminate artifacts and noise, ensuring the integrity and reliability of the results^1,2,3. Over the years, EEG has been invaluable in providing non-invasive and high-temporal resolution measurements of brain activity, making it indispensable for studying the dynamic processes of the brain^4,5. However, the raw EEG signals are often contaminated by a variety of noise sources that can obscure the neural signals of interest, thereby complicating the interpretation of the data⁶.

EEG signals are highly susceptible to contamination from sources such as muscle activity, eye movements, and electrical interference⁷. These artifacts can obscure neural signals, making preprocessing a critical step in EEG analysis. This process typically involves multiple stages-file format conversion, downsampling, filtering, artifact rejection, independent component analysis (ICA), and exclusion of noise-related components^8,9. Each step contributes to improving signal quality and ensuring that the data accurately reflect neural activity¹⁰. The complexity of these processes necessitates the use of sophisticated algorithms and software tools to ensure that the preprocessing is both efficient and effective¹¹.

One of the primary challenges in EEG data preprocessing is the variability in the data, which can differ significantly between subjects and recording sessions^10,12. This variability arises due to differences in the physiological and anatomical characteristics of the subjects, as well as variations in the experimental conditions and equipment used. Additionally, the lack of standardized preprocessing protocols can lead to inconsistencies in data analysis and interpretation^13,14. While several pipelines and preprocessing scripts are available, many of them are either tailored to specific datasets or lack comprehensive documentation, making them less accessible to the broader research community¹⁴. Platforms like GitHub host numerous EEG preprocessing scripts and plugins, facilitating the sharing and collaborative improvement of these tools¹⁵. However, the fragmented nature of these resources underscores the need for a robust, efficient, and standardized editable preprocessing pipeline that can be broadly applied across different datasets and research contexts¹⁶. Such a pipeline would not only enhance the reproducibility of EEG studies but also provide a foundation upon which the research community can build and adapt for specific needs¹⁷.

In addition to EEGLAB, which forms the basis of PIPEMAT-RS due to its extensive user base and integration with MATLAB, other EEG processing toolboxes are also widely used in the field. For instance, MNE-Python is an open-source Python-based platform that offers advanced functionalities for source localization, sensor-space analyses, and integration with machine learning workflows. FieldTrip, a MATLAB-based toolbox like EEGLAB, is known for its flexible scripting environment and comprehensive support for time-frequency and connectivity analyses. While these platforms provide powerful alternatives, they typically require steeper learning curves and more advanced programming skills. PIPEMAT-RS was designed to bridge accessibility and transparency, especially for researchers seeking an editable and modular solution for resting-state EEG preprocessing using EEGLAB. Nonetheless, the conceptual structure of PIPEMAT-RS can be adapted to other toolboxes as needed, fostering interoperability and customization across different software environments.

Several automated EEG preprocessing pipelines have been developed in recent years, such as RELAX¹⁸, Automagic¹⁹, APP²⁰, and PREP⁷. These pipelines incorporate advanced algorithms for artifact removal and data cleaning, often employing ICA-based strategies with limited transparency or customization for non-expert users. For instance, RELAX introduces a multi-step ICA cleaning method that improves artifact rejection but relies on complex configurations that may not be easily modifiable²¹. In contrast, PIPEMAT-RS is designed to provide a flexible, transparent, and educationally accessible alternative, enabling researchers to understand and manually adjust each preprocessing step. This transparency allows PIPEMAT-RS to be used as both a research and teaching tool while maintaining rigor and reproducibility. Moreover, PIPEMAT-RS emphasizes documentation, modular execution, and adaptability across datasets-features that are often underreported or absent in other pipelines.

Economically, the development and implementation of a standardized editable preprocessing pipeline for EEG data can have significant impacts. The global EEG devices market was valued at approximately USD 1.2 billion in 2020 and is expected to grow at a compound annual growth rate (CAGR) of 7.5% from 2021 to 2028²². Efficient preprocessing pipelines, particularly those that are editable and modular, can reduce the time and cost associated with data analysis-an important consideration in large-scale clinical trials and public health research. By minimizing human error and automating repetitive tasks, such pipelines can improve the accuracy of EEG data analysis and lead to more reliable research outcomes¹³. This, in turn, can accelerate the development of new diagnostic tools and therapeutic interventions, potentially reducing healthcare costs and improving patient outcomes¹.

This paper aims to present and validate a standardized editable Preprocessing Integrated Pipeline for Resting-State EEG using MATLAB (PIPEMAT-RS). Designed to be both robust and efficient, this editable pipeline is versatile and can be widely applied to various datasets and research scenarios. The seven comprehensive steps include file format conversion, downsampling, filtering, artifact rejection, independent component analysis (ICA), and noise related components exclusion, all integrated into a cohesive workflow^23,24. The development of PIPEMAT-RS was guided by the need to address common challenges in EEG preprocessing and to provide a tool that researchers can use to streamline their data analysis processes²⁵. By integrating with commonly adopted practices in the field and leveraging the capabilities of MATLAB, PIPEMAT-RS aims to offer a reliable, editable, and user-friendly framework to support EEG data preprocessing²⁶.

To validate the effectiveness of PIPEMAT-RS, we have applied it in several studies. In one study, PIPEMAT-RS was utilized to investigate EEG biomarkers in stroke patients, identifying significant correlations between maladaptive motor function and depressive profiles²⁷. Another study applied PIPEMAT-RS to analyze neural signatures of brain compensation in stroke patients using EEG and TMS, revealing critical insights into the brain's adaptive mechanisms post-stroke²⁸. Further validation comes from a study that explored the neural adaptations and compensations in fibromyalgia patients²⁹. Additionally, another study demonstrated how delta and theta bands in resting-state EEG serve as compensatory mechanisms in chronic neuropathic pain³⁰.

These studies collectively highlight PIPEMAT-RS's utility in processing and analyzing EEG data, establishing its efficacy across various conditions and applications. By providing a standardized approach, this pipeline not only enhances the reproducibility of EEG studies but also facilitates the broader adoption of consistent preprocessing methods in the research community. Through this paper, we aim to contribute a valuable resource to the field, promoting more accurate and reliable EEG data analysis^16,17. In doing so, we hope to support the advancement of research in neurophysiology and the development of novel therapeutic approaches for neurological and psychiatric conditions.

The foundational preprocessing steps used in PIPEMAT-RS are rooted in robust, widely validated methods that have become essential in EEG research for ensuring data quality and reproducibility^2,8,14. Although these methodologies are well-established, they remain crucial due to their effectiveness in addressing common EEG artifacts and noise, which continues to support reliable analyses across studies. However, few studies have recently provided a fully documented, standardized pipeline tailored for resting-state EEG. By integrating these proven techniques with recent advances, such as ICLabel for automated artifact classification, PIPEMAT-RS delivers a contemporary, structured approach that enhances replicability and accessibility for researchers, fulfilling an ongoing need for rigorously documented preprocessing workflows in the field.

While PIPEMAT-RS pipeline follows established preprocessing steps commonly employed in EEG studies, its documentation in the form of a scientific paper fills an important gap in the literature. Many EEG studies rely on similar preprocessing methods but often lack a comprehensive, step-by-step description that allows for full reproducibility and ease of adoption by other researchers. By detailing each component of PIPEMAT-RS and presenting a standardized sequence for resting-state EEG analysis, this work aims to enhance transparency, minimize user variability, and provide a replicable framework applicable across studies. Rather than introducing novel methods, this contribution is intended to serve as an instructional and practical guide that helps standardize preprocessing practices and improve accessibility for a wide range of EEG researchers, particularly those working with resting-state data.

The preprocessing of EEG data is a critical step that significantly influences the quality and reliability of subsequent analyses. The Preprocessing Integrated Pipeline for Resting-State EEG using MATLAB (PIPEMAT-RS) was developed to address common challenges in EEG data preprocessing by providing a comprehensive, standardized approach applicable to various datasets. This structured, editable pipeline consists of seven key steps: i) File Format Conversion; ii) EEG Montage; iii) Downsampling and Filtering; iv) Artifact Rejection and Rereferencing; v) Independent Component Analysis (ICA); vi) ICLabel Classification; vii) Data Normalization. Each step is meticulously designed to enhance signal quality and facilitate the extraction of meaningful information from raw EEG data.

A critical feature of PIPEMAT-RS is that each preprocessing step concludes with saving the dataset under a distinct filename. This structure ensures that, for each processed file, a derivative version corresponding to each specific step is generated. This systematic approach allows users to access data at any point within the seven-step preprocessing pipeline. The saved files include identifiers that clearly indicate the stage of preprocessing, making it easier to track and manage data throughout the workflow.

PIPEMAT-RS begins with file format conversion, where raw EEG data files are transformed into a MATLAB-compatible format. This conversion is essential for the seamless handling of data in MATLAB's robust computational environment. Following this, the EEG montage step assigns precise spatial locations to each electrode on the scalp based on standardized electrode placement systems. This spatial information is crucial for the accurate interpretation of EEG data in neurophysiological studies.

Next, the pipeline applies downsampling and filtering to reduce the data's sampling rate and eliminate noise while retaining the frequency components relevant to neurophysiological research. Downsampling decreases the computational load and storage requirements without compromising data quality. Filtering removes specific frequency components known to be associated with artifacts, thereby preserving the essential neural signals.

Subsequently, artifact rejection and rereferencing are performed. This step identifies and removes segments of data contaminated by noise sources such as eye blinks, muscle activity, and electrode movement. Automatic artifact rejection algorithms are utilized to minimize manual intervention, reducing the potential for human error and increasing preprocessing efficiency. Rereferencing the data to the average of all electrodes helps to mitigate the influence of any single electrode and provides a stable reference for the EEG signals.

The Independent Component Analysis (ICA) step further refines the data by separating mixed signals into their independent sources. This decomposition facilitates the identification and removal of artifacts such as eye blinks, muscle artifacts, and line noise. ICA is a powerful technique that ensures the remaining data accurately reflects underlying neural activity.

Finally, ICLabel classification is applied to automatically identify and remove components classified as artifacts. ICLabel assigns probabilistic labels to independent components, categorizing them as brain activity, muscle activity, eye blinks, heartbeats, line noise, or channel noise. Components with a high probability of representing brain activity are retained, while those identified as artifacts are removed. This automated approach significantly reduces manual effort and ensures consistent and objective classification across datasets.

Each of these preprocessing steps is essential for improving the quality of EEG data. By standardizing the preprocessing workflow, PIPEMAT-RS minimizes variability introduced by different preprocessing methods, facilitating the comparison of results across studies. The pipeline's implementation in MATLAB, a widely-used platform in the neuroscience community, ensures accessibility and ease of integration into existing research workflows.

The development of PIPEMAT-RS was guided by commonly adopted practices in EEG data analysis, incorporating methods and techniques validated in previous research. Each step was carefully designed and rigorously tested to ensure its effectiveness in enhancing data quality (Figure 1). This comprehensive approach not only improves the reliability of EEG data but also supports the identification of neurophysiological markers critical for understanding various neurological and psychiatric conditions.

Protocol

PIPEMAT-RS was developed as part of our ongoing research project, approved by the Ethics Committee of the Clinics Hospital, University of São Paulo Medical School (CAAE: 86832518.7.0000.0068). This project aims to enhance the quality and reliability of resting-state EEG data for various neurological and psychiatric studies. The data used in this study are derived from a cohort study, the protocol of which was previously published by Simis et al. (2021)³¹. The main study is still in progress, with data collection ongoing for the remaining clinical groups. The script for the pipeline is provided in Supplementary File 1.

1. File format conversion

1. Prepare the data directory.
   1. Open MATLAB and navigate to the directory containing raw EEG data files.
   2. Define the path using pathname = 'C:\\Path\\To\\RawData\\'; and list subjects with name_subj_raw = {'subject1', 'subject2', ...};.
   3. If the dataset is already in .mat format, skip this step and proceed to EEG Montage.
2. Convert files.
   1. Load each raw data file using the appropriate MATLAB function for the file type. For example, use load for .mat files or file-specific EEGLAB functions such as pop_biosig (for .edf, .bdf) or pop_fileio (for various proprietary formats).
      NOTE: Users should adapt this step based on the format of their raw EEG files and refer to EEGLAB's documentation for additional file support.
   2. Save each file in .mat format, ensuring data integrity and retaining all relevant metadata.

2. EEG montage

1. Load EEG data.
   1. Open MATLAB and navigate to the directory with the .mat files.
   2. Use the pop_loadset function from the EEGLAB toolbox to load each subject's EEG dataset (e.g., EEG = pop_loadset('filename.set');).
2. Apply electrode locations.
   1. Ensure the appropriate electrode location file (e.g., EEGSystem_128.loc) corresponding to the EEG cap configuration is available.
   2. Use the pop_chanedit function to apply electrode locations. For example:
      EEG = pop_chanedit(EEG, 'load', {'EEGSystem_128.loc', 'filetype', 'loc'});
   3. Visually confirm that the electrode positions were loaded correctly by reviewing the channel layout using EEGLAB's GUI.
3. Save the dataset.
   1. Save the updated EEG dataset using the pop_saveset function (e.g., pop_saveset(EEG, 'filename', 'subject1_loc.set')).

3. Downsampling and filtering

1. Downsample the EEG Data.
   1. Reduce the sampling rate of the EEG data to 250 Hz using the pop_resample function (optional).
      NOTE: The pop_resample function automatically applies an appropriate low-pass anti-aliasing filter before downsampling. This step ensures that high-frequency components are attenuated prior to resampling, preventing aliasing artifacts and preserving the quality of the EEG signal.
2. Apply band-pass filtering.
   1. Apply a band-pass filter from 1 Hz to 50 Hz to channels 1-64 using the pop_basicfilter function with a Butterworth design and filter order of 2.
   2. Use zero-phase forward and reverse filtering to prevent phase distortion.
3. Apply notch filtering (Optional).
   1. Apply a notch filter at 50 or 60 Hz to channels 1-64 using pop_basicfilter with the 'PMnotch' design and filter order of 180.
      NOTE: This step is optional and should be used when analyzing oscillatory activity above 30 Hz (e.g., gamma band). In such cases, researchers may increase the upper limit of the band-pass filter (e.g., up to 80 or 100 Hz), making the notch filter crucial for eliminating powerline interference. The command is provided in the pipeline for convenience and can be easily adjusted depending on the local electrical frequency (e.g., 50 Hz or 60 Hz).

4. Artifact rejection and rereferencing

1. Perform automatic artifact rejection.
   1. Use the clean_rawdata function with parameters (5, [-1], 0.7, -1, 8, 0.85) to automatically detect and remove flatline channels, noisy segments, and low-frequency drifts. These parameters correspond respectively to:
      FlatlineCriterion (5): removes channels flat for more than 5 s.
      ChannelCriterion ([-1]): disables channel correlation-based rejection.
      LineNoiseCriterion (0.7): removes channels with excessive line noise above this threshold.
      Highpass (−1): disables additional high-pass filtering during cleaning.
      BurstCriterion (8): removes data bursts that exceed 8 standard deviations.
      WindowCriterion (0.85): requires 85% of data in a window to be clean to retain it.
      NOTE: Users may adjust these values depending on dataset characteristics. For instance, reducing the BurstCriterion will result in more aggressive artifact rejection, while lowering the WindowCriterion will increase the tolerance for noisy data windows.
2. Perform manual artifact inspection (optional).
   1. If necessary, visually inspect the EEG data using eegplot from EEGLAB to identify residual artifacts not detected by automatic methods.
   2. Manually mark and remove remaining artifacts to ensure high data quality.
      NOTE: This step is optional and primarily recommended when dealing with datasets that present non-standard noise patterns or when greater precision is needed beyond automated procedures.
3. Rereference the data.
   1. Use the pop_reref function to rereference EEG signals to the average of all electrodes, ensuring that the original reference electrode is retained in the dataset.
      NOTE: In this pipeline, average re-referencing is performed using the available channels without reconstructing the original reference electrode. However, users aiming for enhanced rank consistency and cross-subject comparability may consider temporarily reinserting the original reference electrode (e.g., Cz) as a zero-valued time series before re-referencing. This allows the reference to contribute to the average calculation. Afterward, the added channel can be removed to avoid rank deficiency before performing ICA. This process is exemplified in the RELAX_average_rereference.m function of the RELAX pipeline, which also includes interpolation of all channels prior to referencing. While PIPEMAT-RS maintains a simplified version for clarity and accessibility, users are encouraged to adapt this step as needed depending on the goals of their analysis.

5. Independent component analysis (ICA)

1. Load preprocessed EEG data.
   1. Load the file with the suffix _loc_filt_cleanraw_reref.set using the pop_loadset function.
2. Perform ICA.
   1. Run ICA using the pop_runica function with the runica algorithm to decompose the data into independent components.The 'runica' algorithm, based on the Infomax approach, maximizes the statistical independence of the components and is widely used in EEG preprocessing.
      NOTE: For users seeking faster ICA computation, alternative algorithms such as picard may be used by specifying icatype, picard in the function call. These alternatives offer comparable performance and may reduce computation time.
3. Save the ICA dataset.
   1. Save the dataset with the suffix _loc_filt_cleanraw_reref_ICA using pop_saveset to indicate that ICA has been performed.

6. ICLabel classification

1. Load the ICA dataset.
   1. Import the dataset with ICA components using the pop_loadset function.
2. Apply ICLabel classification.
   1. Run the pop_iclabel function with the default model to classify independent components into brain and artifact categories.
   2. Identify components with a probability greater than 0.7 of representing brain activity.
3. Remove artifact components.
   1. Use the pop_subcomp function to remove all components with ICLabel brain probability below 0.7.
   2. Retain only components with a brain activity probability above 0.7. The threshold of 0.7 balances the need to preserve genuine neural signals while ensuring the effective removal of artifacts.
4. Save the cleaned dataset.
   1. Save the cleaned dataset with the suffix _loc_filt_cleanraw_reref_ICA_Cleaned using pop_saveset to indicate the removal of artifact components.

7. Data normalization

1. Load cleaned EEG data.
   1. Import the dataset with artifact-free ICA components using the pop_loadset function.
2. Apply Z-Score normalization.
   1. Normalize each EEG channel using Z-score transformation by computing (x - mean(x)) / std(x) for each channel individually.
      NOTE: Z-score normalization is optional and should be applied based on the analytical goals of the study. While it reduces inter-subject variability and standardizes data for group-level comparisons, it may also attenuate topographical distinctions and eliminate meaningful amplitude differences across channels. Researchers interested in preserving raw signal magnitude or spatial distribution are advised to skip this step.
3. Save the final dataset.
   1. Save the fully preprocessed and normalized dataset with the suffix _loc_filt_cleanraw_reref_ICA_Normalized using pop_saveset to indicate the completion of preprocessing.

Results

The validation of the EEG preprocessing pipeline is essential to ensure the reliability and effectiveness of the methods employed for extracting meaningful neurophysiological data. This process involves empirical evidence from published studies, statistical assessments, and comparisons with established benchmarks in the field.

The robustness and utility of PIPEMAT-RS have been demonstrated through its successful application in several published studies spanning a range of neurological and psyc...

Discussion

The PIPEMAT-RS pipeline was developed to provide a standardized, efficient method for preprocessing resting-state EEG data. Critical steps in this protocol include artifact rejection and Independent Component Analysis (ICA), both of which significantly enhance the signal-to-noise ratio and ensure the extraction of meaningful neural signals. The combination of automatic artifact rejection using the clean_rawdata function and manual inspection via eegplot ensures comprehensive artifact management, balancing efficiency and ...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
MATLAB	MathWorks	R2020a or newer	Required for executing the PIPEMAT-RS script.
EEG Recording System	Brain Products	https://www.brainproducts.com/	EEG acquisition system used for data collection. Other brands  include BioSemi / ANT Neuro (or equivalent)
EEGLAB Toolbox	Swartz Center for Computational Neuroscience	N/A	Open-source MATLAB toolbox for EEG analysis.
Electrodes (Ag/AgCl)	Brain Products	https://www.brainproducts.com/solutions/r-net/	Used in EEG data acquisition. Other brands  include BioSemi / ANT Neuro (or equivalent)
ICLabel Plugin	Swartz Center for Computational Neuroscience	N/A	Automated artifact classification tool for EEG.
PIPEMAT-RS Script	N/A	N/A	Custom MATLAB script for standardized EEG preprocessing.
Signal Amplifier	Brain Products	https://www.brainproducts.com/solutions/#amplifiers	Amplifies EEG signals for processing. Other brands  include BioSemi / ANT Neuro (or equivalent)
Standard 64-Channel EEG Cap	Brain Products	https://www.brainproducts.com/solutions/#electrodes-caps	Electrode cap for EEG recording. Other brands  include BioSemi / ANT Neuro (or equivalent)