Real time identification of μ wave with wavelet neural networks

In the rehabilitation of paralyzed patients, the functional electrical stimulation (FES) or prostheses is often adopted in clinical practice. One of the key issues in these new technologies is the source for generating control commands. The brain computer interface (BCI) creates an alternative pathway from the brain potentials. In this investigation, we construct real-time system to percept the voluntary movement of right thumb as a basic study of BCI. We combine the wavelet transformation and neural network as Wavelet Neural Network (WNN) identify the attempt of voluntary thumb movement. Three types of classification methods: realtime classification without network update, real-time classification with update and conevntional power spectral analyses are compared, and it was found that the WNN with off-line retraining shows better successful rate up to 80%.