Método de classificação de sinais de eletroencefalograma para auxílio ao diagnóstico do transtorno do espectro autista

Abstract

Recent research indicates a growing number of children diagnosed with autism spectrum disorder (ASD), a disease characterized by symptoms that directly impact the fields of behavior, communication and social interaction. According to the Centers for Disease Control and Prevention (CDC), 1 in every 36 8-year-old children are autistic in the United States, a number 22% higher than in the previous survey in 2021. These facts have driven the development of new tools to the diagnosis of the disease, which is essentially clinical (behavioral observation and interview), since there is no requirement for a specific exam. The present work explores an approach with CNN and RNN, combined with signal pre-processing techniques, such as ordering by percentage difference, with the aim of assisting in the diagnosis of ASD, based on the classification of electroencephalogram (EEG) signals. Using the database of studies by Milne et al., from the University of Sheffield, it is shown that the best proposed model achieved a classification accuracy of 99, 63%, which corroborates the system’s ability to distinguish between individuals with ASD and typically developed.