Background: The electroencephalography (EEG) signals are known to involve the\r\nfirings of neurons in the brain. The P300 wave is a high potential caused by an\r\nevent-related stimulus. The detection of P300s included in the measured EEG signals\r\nis widely investigated. The difficulties in detecting them are that they are mixed with\r\nother signals generated over a large brain area and their amplitudes are very small\r\ndue to the distance and resistivity differences in their transmittance.\r\nMethods: A novel real-time feature extraction method for detecting P300 waves by\r\ncombining an adaptive nonlinear principal component analysis (ANPCA) and a\r\nmultilayer neural network is proposed. The measured EEG signals are first filtered\r\nusing a sixth-order band-pass filter with cut-off frequencies of 1 Hz and 12 Hz. The\r\nproposed ANPCA scheme consists of four steps: pre-separation, whitening,\r\nseparation, and estimation. In the experiment, four different inter-stimulus intervals\r\n(ISIs) are utilized: 325 ms, 350 ms, 375 ms, and 400 ms.\r\nResults: The developed multi-stage principal component analysis method applied at\r\nthe pre-separation step has reduced the external noises and artifacts significantly.\r\nThe introduced adaptive law in the whitening step has made the subsequent\r\nalgorithm in the separation step to converge fast. The separation performance index\r\nhas varied from -20 dB to -33 dB due to randomness of source signals. The\r\nrobustness of the ANPCA against background noises has been evaluated by\r\ncomparing the separation performance indices of the ANPCA with four algorithms\r\n(NPCA, NSS-JD, JADE, and SOBI), in which the ANPCA algorithm demonstrated the\r\nshortest iteration time with performance index about 0.03. Upon this, it is asserted\r\nthat the ANPCA algorithm successfully separates mixed source signals.\r\nConclusions: The independent components produced from the observed data using\r\nthe proposed method illustrated that the extracted signals were clearly the P300\r\ncomponents elicited by task-related stimuli. The experiment using 350 ms ISI showed\r\nthe best performance. Since the proposed method does not use down-sampling and\r\naveraging, it can be used as a viable tool for real-time clinical applications.
Loading....