Deep Learning Applied to ERP in the Search for Spatial Intelligence Signatures

Abstract

Neuroscience every day seeks to better describe human behavior by more assertive solutions to problems such as classification in EEG data and for that techniques, such as deep learning, have been widely used, making it feasible to implement models for the automatic processing of biosignals. In this sense, the present work proposes to apply deep learning to handle time series of visual evoked potentials, with the objective to observe the presence of patterns in ERP signatures to associate them with the degree of spatial intelligence of a subject, from the region frontal and parietal of the cerebral cortex, as an alternative to traditional task-based IQ tests, given the problem of lack of automatic intelligence signature pattern recognition systems, as a behavioral and electrophysiological biomarker associated with visual-spatial ability. Using the technique of supervised deep learning, through artificial neural networks (deep feedforward) MLP type for feature recognition in intellectually gifted subjects compared to control subjects collected while performing a classic three-dimensional mental rotation task. Contributing directly to the area of neuroscience, by confirming the hypothesis that in a series of visual ERPs there is a signature pattern of a subject's spatial intelligence and, therefore, its automatic classification is possible, and filling gaps in the literature regarding automatic extraction of intelligence biomarkers.