Classificação de sinais cerebrais sob a ótica dos Large Language Models aplicação na reabilitação clínica

Abstract

The recent advancements in natural language processing(NLP) have introduced novel artificial intelligence models fordata classification, extending their scope to analyzing brainsignals acquired via electroencephalogram (EEG). Amongthese developments, the transformer architecture, which hasbecome available in recent years, has provided researcherswith a powerful model to explore and evaluate its capabilitiesin various EEG-related studies, including developingnew assistive devices tailored for individuals with impairedcommunication skills. This work leverages the transformermodel to classify P300 event-related potentials on publiclyavailable EEG data, aiming to benchmark its accuracy againstestablished algorithms documented in literature. Upon conductingthe case study, the results reveal that the transformerachieves a noteworthy accuracy rate of 95%, indicating itsviability as a classifier for P300-based spellers.