Impact of Ocular Artifact Removal on EEG-Based Color Classification for Locked-In Syndrome BCI Communication

Locked-in Syndrome (LIS) is a neurological condition that results in paralysis of the body and the loss of communication abilities while leaving the patient’s cognitive function unaffected [1]. This can significantly impact their quality of life, as previously simple tasks become impossible. Developing a communication system based on electroencephalogram (EEG) signals might, therefore, improve the quality of life of the affected. However, analyzing these signals can be challenging due to artifacts, especially ocular artifacts (OAs) from blinking [2]. This study investigates how four different OA removal techniques - Artifact Subspace Reconstruction (ASR), Independent Component Analysis (ICA), Signal-Space Projection (SSP), and a modified version of SSP - affect the accuracy of the convolutional neural networks (CNN) EEGNet and EEGNeX in classifying RGB color exposure and color exposure against a rest-state. The results show that EEGNeX outperforms EEGNet, improving classification by 4–9% depending on the OA removal technique used. While the OA removal techniques do not significantly differ, EEGNeX performs best with accuracies of over 71% for RGB classification and over 85% for Color/Rest classification. These results suggest that RGB color exposure elicits unique EEG patterns in the brain that could be used to develop a general communication model for LIS patients.