Comparative evaluation of probabilistic neural network versus support vector machines classifiers in discriminating ERP signals of depressive patients from healthy controls

This paper describes the design of classification system capable of distinguishing patients with depression from normal controls by event-related potential (ERP) signals using the P600 component. Clinical material comprised twenty-five patients with depression and an equal number of gender and aged-matched healthy controls. All subjects were evaluated by a computerized version of the digit span Wechsler test. EEC activity was recorded from 15 scalp electrodes and recordings were digitized for further computer processing. Features related to the shape of the waveform were generated using a dedicated custom software interface system developed in C++ for the purposes of this work. A software classification system was designed, consisting of (a) two classifiers, the probabilistic neural network (PNN) and the support vector machines (SVM), (b) two routines for feature reduction and feature selection, and (c) an overall system evaluation routine, comprising the exhaustive search and the leave-one-out methods. Highest classification accuracies achieved were 92% for the PNN and 96% for the SVM, using the 'latency/amplitude ratio' and 'peak-to-peak slope' two-feature combination. In conclusion, employing computer-based pattern recognition techniques with features not easily evaluated by the clinician, patients with depression could be distinguished from healthy subjects with high accuracy.