TL;DR
This paper introduces a new, open-source MATLAB toolbox for EEG microstate analysis, aiming to demystify and standardize this powerful technique for understanding brain activity. By providing a transparent, GUI-driven, and scriptable workflow, the toolbox empowers both new and experienced researchers to explore and advance microstate analysis.
What Problem Does This Paper Address?
EEG microstate analysis offers a unique way to characterize brain activity by identifying recurring, short-lived spatial patterns (microstates) in the EEG signal. While conceptually straightforward and neurophysiologically interpretable, the practical application of microstate analysis has been hindered by a lack of transparency in existing methods and toolboxes. This opacity makes it difficult for researchers to fully understand the underlying algorithms, reproduce results, and develop new approaches. The authors aim to bridge this gap by providing a comprehensive, open-source toolbox that makes microstate analysis more accessible, reproducible, and extensible.
Methods
The core of this work is the development of an EEGlab toolbox for MATLAB. The toolbox implements several established microstate analysis techniques, including:
- Clustering Algorithms: K-means, modified K-means (which accounts for polarity invariance and models microstate activations), and Topographic Atomize and Agglomerate Hierarchical Clustering (TAAHC).
- Number of Microstates Selection: Various measures of fit are provided, such as Global Explained Variance (GEV), Cross-validation criterion (CV), Dispersion (W), and Krzanowski-Lai (KL) criteria, to help users determine the optimal number of microstates.
- Backfitting: A method to assign microstate labels to EEG data based on similarity to pre-defined microstate prototypes.
- Temporal Smoothing: Techniques like windowed smoothing and small segment rejection are included to refine microstate label sequences and reduce noise artifacts.
- Microstate Statistics: Calculation of various statistical parameters (e.g., duration, occurrence, coverage, GEV, spatial correlation) to characterize microstate dynamics.
The toolbox offers both a graphical user interface (GUI) for ease of use and a scriptable framework for automating analyses and developing new methods.
Datasets
The paper utilizes a small, publicly available dataset of spontaneous EEG data from four subjects for demonstration purposes. The authors provide a link to download this dataset and the toolbox. No new benchmarks are introduced; rather, the focus is on providing a tool to analyze existing datasets.
Key Insights and Ideas
- Transparency is Key: The authors strongly emphasize the need for transparency in microstate analysis methods to ensure reproducibility and facilitate scientific progress.
- GUI and Scripting Synergy: The toolbox effectively combines a user-friendly GUI for initial exploration with a powerful scripting interface for advanced users and automation, catering to a broad audience.
- Modularity for Innovation: The open-source nature and well-documented code encourage users to modify existing algorithms or implement new ones, fostering innovation in microstate analysis.
- The Importance of Settings: The paper highlights how seemingly small changes in algorithm settings (e.g., number of restarts, convergence thresholds) can lead to substantial differences in microstate segmentation, underscoring the need for careful parameter selection and reporting.
Key Results
The paper primarily serves as an introduction to the toolbox and a guide to microstate analysis methods. The "results" are demonstrated through a tutorial walk-through of the toolbox's functionality. The key outcome is the provision of a functional, open-source toolbox that simplifies and standardizes EEG microstate analysis. The paper also provides a comprehensive overview of commonly used microstate methods, their theoretical underpinnings, and practical considerations.
Limitations and Future Work
- Stochasticity in Algorithms: Some algorithms, like K-means, are stochastic, meaning they can produce different results on repeated runs. The toolbox addresses this with multiple restarts, but users need to be aware of this variability.
- Polarity Invariance: While some methods (modified K-means, TAAHC) are polarity-invariant, others (like the KL criterion) are not, which can be a consideration when choosing methods.
- Empirical Validation: The authors acknowledge that their workflow choices are based on experience and that empirical evidence for optimal settings is lacking.
- Future Work: The authors express interest in advancing methods for selecting the optimal number of microstates and encourage community contributions for further development and new clustering algorithms.
Summary updated