High Performance EEG Analysis for Brain Interface

A successful brain interface (BI) system enables individuals with severe motor disabilities to control objects in their environment (such as a light switch, neural prosthesis or computer) by using Dilly their braia sigaals. SlK" a system IReastlra specific feat1lra of a IJe""'S bnUt sipal dlat relate to his or her intent to affect control, then translates them into coatnli siluk that are used to coatrol a device. ReceBtty. successful applications of the discrete wavelet transform have beep.report.d.~b.rain ili!erface"(Bl) systelDswith one or two EEG channels. For a mufti-channel BI system, however, the lUg" di_easionality of the generated wavelet featares space poses a challenging problem. In this paper, a feature selection method that effectively rcdJlus ~.JJi~eDSiooality of the feature space of a m~han~el, self-paced HI system is _DrQ~ed. Tlte prnll'QKd~_QSeS a two-sq~ feature _ ----·-seleetion-s~*,'_est slIit.ble IIIOtelllC1lt related poUotial features from. the. featurespau. The fim--stage employs mutual information to f"dterout the least discriminant features, resulting in a reduced feature space. Then a genetic algorithm is applied to the reduced feature space to further redllce Widim.ensionality and sekct th.ebest set offcatuRs.. An oIDiae ualysB of tile EEG siguls (18 bipolar EEG cluutnels) of four able-bodied s••ojects showed that the proposed method acqaira low false positive rates at a reasoubly ItiP tne positive rate. The results also sltow that reatares selected fro•• different channels varied considerably from one subject to anotller. The proposed bybrid method effectively reduces the higb dimensionality of the feature space. The variability in features among subjects indicates tbat a user-customized BI system needs to be developed for individual users.