Electromagnetic signals in noninvasive brain-computer interfaces

A brain-computer interface (BCI) translates task-related brain activity into computer commands. Detecting this activity is difficult, as the measured brain signals are generated by multiple sources and also include task-irrelevant brain activity. Using conventional methods such as signal averaging is not possible, because subjects should receive online feedback of their performance. BCI users usually either learn to control some components of their brain activity with the help of feedback or are presented with some stimuli that produce detectable signals in the brain. This thesis reviews BCI research, basic principles of electroencephalography (EEG) and magnetoencephalography (MEG), the sensorimotor cortex, and then describes experimental BCI studies. The thesis comprises of five publications studying 1) sensorimotor cortical activation for BCI, 2) use of MEG for BCIs, 3) single brain signal trials during (attempted) finger movements for online BCI classification and 4) vibrotactile feedback in comparison to visual feedback. Participants were 45 healthy, 9 tetraplegic, and 3 paraplegic subjects. First our results of tetraplegic subjects show that their 10- and 20-Hz rhythmic activity is more widespread and less contralateral than that of healthy subjects, providing a poorer control signal for two-class movement classification. For separating brain signals during right and left attempted movement, we selected features from the low-frequency bands. Second, our results show that for classification, MEG is not superior to EEG for two-class BCI, despite being a more localised measurement technique. Third, brain signals during finger movements could be classified online with high accuracy after basically no training. However, results from the tetraplegic subjects are much worse than those of the healthy subjects. Fourth, we show that vibrotactile feedback can be used as an alternative feedback channel during training and is especially useful when visual attention is needed for application control. On the basis of these and earlier findings, it is concluded that accurate control of noninvasive BCI is possible but requires some training. Future important research involves more work with motor-disabled patients, especially when testing new signal processing methods. Better performance may also be achieved using different feedback modalities