Yes/No Classification of EEG data from CLIS patients

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Completely Locked-in State (CLIS) patients retain cognitive abilities but are paralyzed, necessitating advanced communication aids.
• Existing Brain-Computer Interface (BCI) systems for CLIS patients require improvement in reliability and feature extraction.

Purpose of the Study:

• To evaluate novel electroencephalogram (EEG) features for classifying brain signals in CLIS patients.
• To assess the usability of these features for developing a more reliable communication system for CLIS individuals.

Main Methods:

• EEG data from four CLIS patients were analyzed during an auditory paradigm task.
• Features included spectral measures (power bands, spectral edge frequencies), complexity (Poincaré plots), and connectivity (coherency, weighted Symbolic Mutual Information - wSMI).
• Classification was performed using Random Forest and Support Vector Machine algorithms, considering two data recording cases.

Main Results:

• Classification accuracy ranged from 50.41% to 67.94% across different cases and methods.
• Weighted Symbolic Mutual Information (wSMI) with a 64 ms time lag yielded the highest classification accuracy.
• Random Forest generally outperformed Support Vector Machine in classifying CLIS patient data.

Conclusions:

• The study demonstrates the potential of EEG complexity and connectivity features for CLIS patient signal classification.
• These findings represent a significant step towards developing more effective EEG-based BCI communication systems for individuals with CLIS.