Design and implementation of a multi-PNN structure for discriminating one-month abstinent heroin addicts from healthy controls using the P600 component of ERP signals

A multi-probabilistic neural network (multi-PNN) classification structure has been designed for distinguishing one-month abstinent heroin addicts from normal controls by means of the Event-Related Potentials' P600 component, selected at 15 scalp leads, elicited under a Working Memory (WM) test. The multi-PNN structure consisted of 15 optimally designed PNN lead-classifiers feeding an end-stage PNN classifier. The multi-PNN structure classified correctly all subjects. When leads were grouped into compartments, highest accuracies were achieved at the frontal (91.7%) and left temporo-central region (86.1%). Highest single-lead precision (86.1%) was found at the P3, C5 and F3 leads. These findings indicate that cognitive function, as represented by P600 during a WM task and explored by the PNN signal processing techniques, may be involved in short-term abstinent heroin addicts. Additionally, these findings indicate that these techniques may significantly facilitate computer-aided analysis of ERPs.