TL;DR
This paper introduces TopoToolbox, a novel open-source MATLAB toolbox that leverages sensor topography from EEG/MEG data to derive psychologically meaningful measures of response similarity, magnitude, and timing. By focusing on sensor space rather than attempting direct source localization, TopoToolbox offers a robust and user-friendly approach to analyzing event-related brain activity, overcoming limitations of traditional waveform analyses and enabling reliable cross-individual comparisons.
What Problem Does This Paper Address?
Analyzing event-related electrophysiological data (EEG/MEG) often involves understanding how brain activity changes across different experimental conditions. Traditional methods, like analyzing waveforms at specific electrodes, can be susceptible to issues like reference channel selection, difficulty in combining data across individuals (especially with MEG), and ambiguity in distinguishing between changes in response magnitude and changes in the underlying neural source distribution. While multivariate analyses are increasingly used, particularly in fMRI, their application in EEG/MEG has been limited by the challenge of interpreting sensor-level data, which represents a mixture of neural sources. This paper addresses the need for a method that can reliably quantify psychologically relevant aspects of EEG/MEG responses, specifically focusing on topographic patterns across sensors, to overcome these limitations and provide more interpretable and comparable results.
Methods
TopoToolbox implements a two-stage analysis:
- Angle Test (Topographic Similarity): This stage quantifies the similarity between topographic patterns of different conditions. It calculates the cosine of the angle between sensor vectors in an n-dimensional sensor space (where n is the number of sensors). A value of 1 indicates perfect similarity, while -1 indicates opposite patterns. This test helps determine if different conditions involve a different distribution of underlying neural sources. To establish a null hypothesis, the experiment is split into halves, and within-condition angles are compared to between-condition angles.
- Projection Test (Response Magnitude): If the angle test indicates sufficient similarity between conditions (i.e., no significant difference in source distribution), the projection test is used to measure response magnitude. This involves projecting the sensor pattern of each condition onto a "template" pattern for that individual. This template is typically a clean response pattern from another condition or time window. By normalizing projections for each individual, this method accounts for individual differences in sensor topography and response timing, providing a reliable measure of how much of the template response each condition evokes. This allows for the comparison of response magnitudes across individuals.
Additionally, the toolbox includes an Angle Dynamics Test for assessing pattern similarity over time, particularly useful when waveform peaks are not well-defined. This test applies the angle measure at every time point to track how topographic patterns evolve.
Datasets
Not discussed in the paper. The paper focuses on the methodology and provides examples using data from previously published experiments ([26] and [2]).
Key Insights and Ideas
- Sensor Space is Powerful: The paper argues that analyzing topographic patterns in sensor space, rather than immediately attempting source localization, can yield psychologically meaningful insights and overcome some limitations of traditional methods.
- Distinguishing Magnitude vs. Distribution: A key contribution is the ability to differentiate between a change in the amount of neural activity (magnitude) and a change in where that activity originates (distribution of neural sources). The angle test addresses distribution, while the projection test addresses magnitude.
- Individual Normalization is Crucial: TopoToolbox's ability to normalize for individual differences in sensor topography and response timing is highlighted as a major advantage, leading to more reliable cross-individual comparisons, especially for MEG data.
- "Clean" Templates for Decontamination: The projection test's reliance on a "clean" template pattern is crucial for decontaminating response magnitude measures by removing overlapping waveform responses orthogonal to the template.
- Dynamic Analysis for Unclear Peaks: The angle dynamics test offers a valuable tool for analyzing temporal aspects of brain responses when traditional peak identification is difficult.
Key Results
- The angle test can reliably determine if topographic patterns between conditions are significantly dissimilar, suggesting different neural source distributions.
- The projection test, when the angle test shows similarity, can reliably quantify response magnitude differences across individuals after normalizing for individual variability.
- The angle dynamics test was successfully used to recover the timing of motor response peaks in a cue-locked epoch, aligning with classically defined RMS peaks, and demonstrating its utility for dynamic analyses.
- An example experiment ([26]) showed that TopoToolbox could distinguish between habituation effects on response magnitude versus changes in neural source distribution, a distinction difficult to make with traditional methods.
Limitations and Future Work
- Inverse Problem: The authors acknowledge the inherent "inverse problem" in EEG/MEG, where multiple source configurations can produce the same topographic pattern. While they argue the chance of this is low in their comparative approach, it remains a theoretical limitation.
- Statistical Power: The paper notes that a failure to find a significant difference with the angle test doesn't definitively rule out a difference (Type II error).
- Validation: Further validation with intracranial EEG is suggested as an ideal way to address limitations related to the inverse problem and null hypothesis testing.
- Future Work: The authors plan to supplement TopoToolbox with Bayesian statistics in future work.
Summary updated