Method for exploratory cluster analysis and visualisation of single-trial ERP ensembles.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Analysis

Background:

• Ensemble averaging of event-related potential (ERP) data assumes identical trials, a premise often invalid.
• A need exists for preliminary data analysis to identify cluster structures in ERP datasets.

Purpose of the Study:

• To develop and validate a comprehensive pipeline for cluster analysis of ERP data.
• To address the limitations of ensemble averaging by identifying underlying data structures.

Main Methods:

• A pipeline utilizing Empirical Mode Decomposition (EMD) for denoising single-trial ERPs.
• A bootstrap-based Stability Index (SI) to determine the optimal number of clusters.
• A Genetic Algorithm (GA) for initial centroid selection, followed by k-means clustering.
• Principal Component Analysis (PCA) for visualizing clustering results.

Main Results:

• The pipeline was validated on simulated data and applied to real experimental ERP data.
• Evidence of 6 distinct clusters was found in a language processing study condition.
• Subject influenced cluster membership, as indicated by a two-way chi-square test.

Conclusions:

• The proposed analysis method operates on denoised single-trials with principled cluster number determination.
• Results are presented via an intuitive visualization scheme.
• Cluster analysis is suggested as a preliminary step prior to ensemble averaging in ERP studies.