Decoding Semantic Categories from EEG Activity in Object-Based Decision Tasks

This study explores the potential of decoding semantic categories from EEG-activity for the use in Silent Speech Brain-Computer Interfaces (BCI). We used object-based decision tasks to evoke conscious semantic processing for five different semantic categories in the participants cerebral cortical structures and implemented different feature extraction and classification methods to evaluate possible setups for semantic category detection in BCIs. All of the tested classification methods exceeded the chance level for training and testing on the data of the individual and even for a cross-subject condition. The best individual accuracy achieved was 84.61% for a Common Spatial Pattern (CSP) feature extraction method and Random Forrest (RF) classifier presented for the first time in a 5-class classification task illustrating the potential of this approach for possible future use in Silent Speech BCIs.