Rejecting Artifacts Based on Identification of Optimal Independent Components in an Electroencephalogram During Cognitive Tasks

Eye-blinks and eye movements contaminate electroencephalogram (EEG) results mainly in the form of artifacts, as they generate misleading neural activities in the EEG data. To manage this phenomenon, we proposed an identification technique for the optimal independent components (ICs) of electrical noise generated by artifacts. Our technique integrates independent component analysis and K-means, which is a machine learning method implemented in our previous study. However, we previously evaluated the performance of the method for only artificial EEG noise superimposed on an eye-blink and eye movement template on a resting EEG. In this study, we evaluated the performance of the technique proposed in our previous study using real EEG data during a cognitive task, namely an implicit association task (IAT), that enables the detection of implicit biases in individual subjects. In this task, images or characters were presented visually, and hence many artifacts associated with an eye-blink and eye movement contaminated the EEG. As a result, the averaging numbers and signal-to-noise ratio in an event-related potential, rejected by the identified ICs, improved in most subjects compared with those obtained using the conventional method. The results showed that the proposed method yielded acceptable levels of performance. The proposed method can be employed for visualizing cognitive tasks, such as an IAT, as well as in an artificial EEG.