Classification of Imagined Motor Tasks for BCI

Electroencephalography (EEG) is a well developed technique used in many clinical and research applications. Continuous improvements on quality of scalp electrodes and front-end amplifiers, and data processing and storage have elevated EEG to a standard non-invasive method for monitoring many brain functions. EEG can also provide a new means for sending messages to the external world which is commonly known as a Brain-Computer Interface (BCI). This paper describes different feature extraction techniques for classification of recorded EEG signals. Time and frequency processing of multichannel EEG recordings during four a priori known mental tasks is presented. The four tasks include imagining the movement of an arm or a leg without the execution of the actual motion. During the recording sessions, the imagined movements are separated with intervals of subject relaxation. Different methods were used for feature extraction and classification of the EEG signals as a base for BCI. The results demonstrate that signals from an untrained subject can be classified successfully. The algorithms can be used to establish a real-time direct connection between mental task activity and external communication. In this regard we view the possibility for extending the neuroplasticity of the brain toward direct control of specifically designed external devices.