ECoG-based brain-computer interface using relative wavelet energy and probabilistic neural network

In this study, a brain-computer interface (BCI) using electrocorticograms (ECoG) is proposed. Feature extraction is an important task that significantly affects the classification results. First, the discrete wavelet transform was applied to ECoG signals from one subject performing imagined movements of either the left small-finger or the tongue. After preprocessing, relative wavelet energy of selected 8 channels were extracted and built 40 dimension feature vector. Then the dimension of feature vector was reduced using principal component analysis (PCA). Finally, probabilistic neural network (PNN) was used to classify. The average classification accuracy rate reached a maximum of 91.8% when spread of radial basis functions was 0.11. The offline analysis results showed that ECoG signals could be used in BCI design, and gave new ideas and methods for feature extraction and classification of imaginary movements in ECoG-based BCI research.