Brain Evoked Response Qualification Using Multi-Set Consensus Clustering: Toward Single-Trial EEG Analysis

Abstract

Objective: Scalp electroencephalogram (EEG) provides a substantial amount of data about information processing in the human brain. In the context of conventional event-related potential (ERP) analysis, it is typically assumed that individual trials for one subject share similar properties and stem from comparable neural sources. However, group-level ERP analysis methods (including cluster analysis) can miss important information about the relevant neural process due to a rough estimation of the brain activities of individual subjects while selecting a fixed time window for all the subjects. Method: We designed a multi-set consensus clustering method to examine cognitive processes at the individual subject level. First, consensus clustering from diverse clustering methods was applied to single-trial EEG epochs of individual subjects. Next, the second level of consensus clustering was applied across the trials of each subject. Afterward, a modified time window determination is applied to identify the ERP of interest of individual subjects. Results: The proposed method was applied to real EEG data from the active visual oddball task experiment to qualify the P3 component. Our findings disclosed that the estimated time windows for individual subjects can provide more precise ERP identification than considering a fixed time window for all subjects. Moreover, based on standardized measurementerror and established bootstrap for single-trial EEG, our assessments revealed suitable stability in the calculated scores for the identified P3 component. Significance: The new method provides a realistic and information-driven understanding of the single trials' contribution towards identifying the ERP of interest in the individual subjects.