| Summary: | People with very severe motor-disabilities have to accept a much-reduced quality of life. Unfortunately, these people cannot use assistive devices as present devices require motor activities.
Brain-computer interfaces (BCIs) provide an alternative means of communication between the brain and assistive devices. There are two types of BCIs, synchronous and self-paced. Self-paced BCIs are more practical as they can be used at any time. The vast majority of existing self-paced BCIs are activated by real, attempted, or imagined movements. Few are activated by mental tasks.
The high false positive rates (FPRs) of existing self-paced BCIs render them impractical. There are no self-paced BCIs based on motor movements that have low FPRs. However, self-paced BCIs with low FPRs based on mental tasks have been proposed. Designing a self-paced mental task-based BCI with a zero or near zero FPR and a reasonable true positive rate (TPR) is the goal of this thesis.
We investigated the feasibility of having a self-paced mental task-based BCI with a zero or near zero FPR. The EEG signals from 6 electrodes of 4 subjects performing 4 mental tasks are used. Features were extracted using autoregressive modeling. Different classifiers were tested. The results were promising in that zero FPRs were obtained. The data used, however, had not been collected in a self-paced paradigm.
We then collected the EEG signals from 29 channels of 4 subjects performing 4 mental tasks in a self-paced paradigm. We evaluated the performance of our BCI using this dataset. It yielded an average TPR of 67.26% and zero FPR.
To make the system practical, we decrease the number of channels from 29 to 7 and 5, using 2 approaches that yield local optimal results. The average TPR is sufficiently high (54.60% and 59.98% for systems with 5 and 7 channels) while the FPRs remain zero.
We also study the effects on the performance, as the segment length is varied. For the 7-channel BCIs, the optimum length is between 1 and 2.5 sec. The average TPR is improved to 63.47%. The FPRs are zero. We also show that our BCIs are robust to artifacts. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|
|---|
