ICA-Based Imagined Conceptual Words Classification on EEG Signals

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Independent Component Analysis (ICA) is conventionally used for eye artifact removal in electroencephalography (EEG).
• This study explores ICA's potential for detecting conceptual brain signals, specifically for danger and information categories.

Purpose of the Study:

• To investigate the efficacy of ICA in identifying and classifying brain-produced signals related to conceptual words (danger and information).
• To compare classification accuracies using different feature extraction methods and machine learning algorithms.

Main Methods:

• EEG signals were recorded from participants performing tasks involving traffic signs and directional arrows.
• Independent Component Analysis (ICA) was applied to detect artifactual and conceptual brain signals.
• Autoregressive (AR)-15 model coefficients with a 2.5s time window were identified as optimal features.
• Linear Discriminant Analysis (LDA) and neural networks were employed for classification.

Main Results:

• ICA successfully differentiated between brain signals for danger and information concepts in 7/8 volunteers.
• Classification accuracies were higher in the right hemisphere for 5/8 volunteers and the left hemisphere for 3/8.
• Linear Discriminant Analysis (LDA) outperformed neural networks, achieving over 60% classification accuracy.
• Optimal feature extraction involved ICA output with AR-15 coefficients and a 2.5s time window.

Conclusions:

• ICA is a viable algorithm for recognizing conceptual word meanings and their neural correlates.
• The findings demonstrate comparable results to other neuroimaging techniques like fMRI for specific cognitive tasks.
• LDA combined with ICA and AR-15 features offers a robust method for brain-computer interfaces and cognitive state analysis.