A Multi-Level Integrated EEG-Channel Selection Method Based on the Lateralization Index.

Area of Science:

• Neuroscience and Biomedical Engineering
• Signal Processing and Machine Learning

Background:

• Optimizing channel selection is crucial for portable Brain-Computer Interface (BCI) technology.
• Reducing electrode count without sacrificing decoding accuracy presents a significant challenge.
• Existing methods often increase computational load and neglect cross-subject channel selection.

Purpose of the Study:

• To introduce a novel Multi-level Integrated EEG-Channel Selection method based on the Lateralization Index (MLI-ECS-LI).
• To enable effective channel selection for cross-task and cross-subject scenarios in Motor Imagery BCI (MI-BCI).

Main Methods:

• The MLI-ECS-LI method was developed, utilizing the lateralization index for important channel identification.
• Time and frequency domain features were extracted from channels selected by MLI-ECS-LI.
• Movement types were classified using Least Squares Support Vector Machine (LSSVM), Random Forest (RF), and Support Vector Machine (SVM).

Main Results:

• The MLI-ECS-LI method demonstrated improved decoding accuracies across various scenarios compared to conventional methods.
• Average accuracy improvements ranged from 2.8% to 9.2% depending on the classifier and scenario (single-task, cross-task, cross-subject).
• Significant performance gains were observed across 21 healthy subjects, highlighting the method's robustness.

Conclusions:

• The proposed MLI-ECS-LI method effectively reduces channel selection while maintaining or improving decoding accuracy.
• This approach enhances the utility and practical applicability of portable MI-BCI systems.
• MLI-ECS-LI shows strong potential for real-world BCI applications by addressing key limitations in channel selection.