Robust EEG separability of subject specific records via clustering and data driven metrics

The criteria used to discriminate electroencephalograms (EEG) between subjects in different mental states, performing clinical trials, or suffering from seizures are found by removing subject specific information to develop robust universal features. Information is lost in pursuit of feature detection that could be used to develop links previously overlooked between subjects. We show that these feature spaces allow models to be built that can be used to cluster subjects. These models will form the basis of an algorithm that can quantify similarity between EEG signals regardless of the state of the subject. The existence of a robust comparison metric would enable applications such as biometrics, neural interfaces, and clinical diagnostic support.