Exploiting Temporal Context in High-Resolution Movement-Related EEG Classification

Abstract

The contribution presents an application of a movement-related EEG temporal development classification which improves the classification score of voluntary movements controlled by closely localized regions of the brain. A dynamic Hidden Markov Model-based (HMM) classifier specifically designed to capture EEG temporal behavior was used. Surprisingly, HMM classifiers are rarely used for BCI design despite of their advantages. Because of this we also experimented with Learning Vector Quantization, Perceptron, and Support Vector Machine classifiers using a feature space which captures the temporal dynamics of the data. The results presented in this work show that HMM achieves the best performance due to an a priori information on physiological behavior of EEG inserted to the HMM classifier. Feature extraction process and problems with classification were analyzed as well. Classification scores of 66.7% – 94.7% were achieved in our experiments.

Keywords

• Brain-Computer Interface
• EEG classification
• electroencephalography
• neural network applications
• Hidden Markov Models