Accurate Classification of Visual Evoked Potential Signals Using Spectral Analysis and Feature Extraction Algorithms Combination

Abstract

In the present era, sophisticated signal processing and artificial intelligence methods represent one of the most valuable and pervasive decision-support systems. These systems leverage all of the information present within a signal, encompassing both temporal and spectral domains. By integrating this valuable feature extraction with the processing speed and precision of computers, these systems enhance the quality of decision-making. One of the most extensive areas of application for these systems is medical diagnosis. In this paper, the principal component analysis (PCA) and Support Vector Machine (SVM) are combined and proposed for efficient extraction of the power spectral and square spectrum density features (PSD and SPSD) of visual evoked potential (VEP) signals. We present in this work a performance comparative study using four combination scenarios, applied for the spectral analysis of VEP. The aim is to investigate correctly discriminating pathological and normal cases of a population, and subsequently determine the accuracy and effectiveness of each proposition. The classification results obtained with computer simulation using a VEP dataset show that the SPSD + PCA + SVM based pathology recognition combination outperforms other classifiers in terms of classification performance. This last method is suitable for its potential clinical application in the context of the neuro-physiological exploration and diagnosis in hospitals.