Aplicações de inteligência artificial no apoio ao diagnóstico do Transtorno do Espectro Autista – revisão sistemática

Abstract

This systematic review aims to analyze the use of artificial intelligence (AI) technologies in supporting the diagnosis of Autism Spectrum Disorder (ASD), focusing on development, applied methods, implementations, and reported results. The methodology was conducted according to the PRISMA protocol, encompassing searches in the Web of Science, Scopus, and PubMed databases, including articles published between January 2021 and March 2025, in English and Portuguese. Information was extracted on the origin of the data (neuroimaging, EEG, eye tracking, genetic data, among others), the algorithms used, the performance metrics (accuracy, sensitivity, specificity, AUC), the validation techniques used, and the presence of explainable AI approaches (XAI). The results demonstrate that AI-based solutions can complement traditional clinical methods, contributing to faster, more standardized, and more accurate diagnoses. However, the consolidation of its widespread use still faces significant barriers, such as a scarcity of reliable databases, a lack of robust clinical validation, and ethical and technical challenges. Thus, this study contributes to systematizing existing knowledge, identifying gaps in the literature, and indicating strategies for future research aimed at strengthening the application of AI in the diagnosis of ASD.