Multichannel Analysis of EEG Signals for Screening Alcoholics

Multi-channel analysis of EEG signals were analyzed in the project to detect alcoholism. A digital signal-processing algorithm that automates the classification of signals as normal or alcoholic is studied here. The method exploits the existing digital signal processing concepst sucah as signal modeling and and spectral estimation for feature extraction and classification. The mult-channel AR modeling and Cepstral theory were used for feature extraction. The EEG signals use in the project include 32 channels recorded from different portions of the brain for three minutes duration. 25 signals of each subject were taken for analysis. A classifier is developed based on Linear Discriminant Analysis (LDA) and Leave-One-Out method (LWO), the signal deatures were classified based on the norm distances to maximize the accuracy. Maximum Likelihood (MLM) or Euclidean distance is used to extract the norm distance between the signal under test and the reference vector. This was repeated for the entire training database. The classifier thus obtained gave the overall accuracy rate of the system. The accuracy rate obtained with AR coefficients is 72% and the accuracy rate with cepstral coefficients is 62%.