Pattern recognition of human brain electrical potentials

Few problems are more challenging than decoding the mass electrical activity of the human brain. Although this goal is distant, there are currently several useful applications of a correlative approach based on pattern recognition. These include investigation of the neural concomitants of higher cognitive functions, diagnostic screening, differential diagnosis and prognostic assessment of neurologic disorders, classification of the stages of ordinary and disturbed sleep, and classification of the neuroelectric patterns associated with different types of psychotropic drugs. Since brain electrical potentials (BEP's) are correlated with a variety of behavioral and clinical variables, especially tight experimental designs are necessary. Irrelevant demographic, behavioral, and pharmacologic factors must be controlled, or else the results will be ambiguous. Primary analysis, which usually consists of spectral analysis, linear prediction, or zero-cross detection, should match the time scale and dynamics of the states or processes being investigated. Nonneural contaminants must be removed from BEP's prior to computation of summary features. Principal components analysis, ad hoc methods, and stepwise discriminant analysis have been used to extract independent, intuitively appealing, and good-classifying features, respectively. If features are not neurophysiologically interpretable, the utility of even successful classification is likely to be limited in most applications. Features should be expressed in a form which is insensitive to irrelevant anatomic and metabolic variables. Within-persons, between-classes standardization is useful in this regard. Sample size determination is application-specific since intra- and interperson BEP variability is different for different problems. Most pattern classification algorithms have been applied to BEP's including decision functions, trainable classification networks, distance functions, syntactic methods, and hybrids of the preceding. Because of its wide availability, most studies have used stepwise linear discriminant analysis. Since the assumption of equal covariance matrices between classes is generally invalid for BEP's, classification performance on the same data set should be consistently higher with algorithms not based on such an assumption. Syntactic methods have been used to mimic the traditional visual assessment of clinical EEG polygraphs for diagnostic screening and prognostic assessment. The most convincing validation of a BEP classification function is its application to an independent sample of persons. Leave-out-TV replications are obviously of questionable value if the total data set was previously used to extract a good-classifying subset of features.