Revisão sistemática de estudos que utilizam sinais de EEG para criação de conjunto de dados aplicáveis a aplicações médicas

Abstract

This work presents a systematic literature review to analyze studies utilizing Electroencephalogram (EEG) signal datasets in medical applications. The objective was to investigate how these databases and the use of machine learning contribute to the recognition and prediction of neurological and psychiatric diseases. The methodology followed the PRISMA guidelines, employing a search strategy in the Scopus and PubMed databases, with a temporal filter from 2022 until October 2025. Forty-five primary studies were selected and evaluated regarding their methodological and technical quality. The results indicate a predominance of research focused on epilepsy, with a shift from detection to seizure prediction. A growing trend in the diagnosis of neurodegenerative diseases, such as Alzheimer's, and psychiatric disorders was also observed. Technologically, the hegemony of Deep Learning techniques was identified, especially Convolutional Neural Networks (CNN) and hybrid architectures, often associated with the conversion of 1D signals into 2D visual representations. It is concluded that, although technical maturity is high, the field faces challenges related to the limited availability of public data for non-epileptic conditions and the need for greater model interpretability to facilitate effective clinical application. The full article related to this work is available at https://doi.org/10.13140/RG.2.2.26201.58723.