Identification of diagnostic evoked response potential segments in Alzheimer's disease.

Evoked response potentials (ERPs) to brief flashes of light were analyzed for constituent features that could be used to distinguish individuals with Alzheimer's disease (AD, n = 15) from matched control subjects (n = 17). Statistical k nearest-neighbor methods distinguished AD from control with a maximum sensitivity of 29% and false alarm rate of 12%. The comparable sensitivity/false-alarm values for a statistical projection pursuit method and an extended projection pursuit method, which selectively identify discriminative features for classification, were 75%/18% and 100%/6%, respectively. The results demonstrate that combinations of selected ERP time segments across different electrodes contain signal features that discriminate AD from control subjects with high sensitivity and specificity.