(17) A bayesian network approach to brain-computer interface using independent component analysis and equivalent current dipole source localization
M.Sakamoto, T.Yamazaki, K.Kamijo, T.Yamanoi
NEUROSCIENCE2010, 2010-11
In this study, we present a new method for developing single-trial-EEG-based BCI (Brain-Computer Interface). This method could discriminate left and right hands to be imagined from the single-trial EEGs measured during the movement imagery tasks. The method consists of the following three steps: (1) independent component analysis (ICA) for each of 32-channel single-trial EEGs; (2) equivalent current dipole source localization (ECDL) for reconstruction of each ICA on the scalp surface; (3) Bayesian network model (BNM) construction using the ECDL results. The obtained BNM involved fifteen nodes corresponding to the brain sites such as frontal, temporal, occipital abd cingulate gyri, hippocampus, insula, left and right parietal cortices, left and right motor ares, left and right cerebellum, left and right somatosensory areas, and others. Each node has a value of one or zero whether ECDs are localized to the brain sites. The present BNM structure was estimated from the ECDL using conditional independence (CI) tests. The classification rule based on the BNM is that, for right-handed subjects, there is a significant difference in node activities (conditional probability) between left and right motor areas during the right-hand-movement imagery, while no difference during the left-hand-movement one. The node activities were inferred by the belief propagation. This hypothesis was statistically validated for four out of five subjects, as shown in Table 1 below, and turned out to be a classification rule for our single-trial-EEG-based BCI.
>>
NEXT