A inteligência artificial no diagnóstico precoce de autismo: uma revisão integrativa

Abstract

INTRODUCTION: Autism Spectrum Disorder (ASD) is characterized by deficits in social communication and interaction, restricted and repetitive patterns of behavior, and differences in sensory processing. Children with ASD may have difficulty understanding the feelings and thoughts of others, react negatively to change, and exhibit social isolation and school difficulties. The prevalence of ASD has increased, impacting child development. Early diagnosis is crucial for interventions that improve the quality of life and development of affected children. Artificial Intelligence (AI) emerges as a possibility for the development of innovative tools, with the potential to improve the early identification of ASD. In this context, this study seeks to understand the main AI tools for the early diagnosis of ASD, evaluating performance metrics. OBJECTIVE: To investigate in national and international scientific production, studies on the validation of Artificial Intelligence tools for the early diagnosis of ASD, with emphasis on the values of sensitivity, specificity, accuracy, and precision. METHOD: This is an Integrative Review whose study question was delimited using the PIRD strategy: Participants/Population (P): children under 10 years old; Test to be evaluated (Artificial Intelligence) (I): AI-based tools; Reference test (R): Conventional tools and/or scales; Diagnosis of interest (D): Early diagnosis of ASD. Thus, the study question was: In children with ASD, what is the diagnostic validity of AI-based tools compared to conventional scales and methods for early diagnosis based on studies identified in the scientific literature. Original studies published between 2018 and 2023 were included, in the databases PubMed, BVS, LILACS, SciELO, Cochrane, Google Scholar, MEDLINE, EMBASE, and Web of Science, in English, Spanish, and Portuguese. The Descriptors in Health Sciences / Medical Subject Headings (DeCS/MeSH) were used, combined by the Boolean operators AND and OR. The analysis of the studies was performed using descriptive evaluation of the publications and of tests of AI-based tools (sensitivity, specificity, accuracy, and precision). RESULTS: Ten studies from different countries were included. AI tools showed high sensitivity, specificity, accuracy, and precision, showing promise in the early identification of ASD, especially through eye tracking and facial expression analysis. One study classified ASD into severity levels (mild, moderate, and severe) using AI. All studies showed a higher prevalence of ASD in males. DISCUSSION: AI tools, such as eye tracking and facial expression analysis, can identify subtle and specific patterns for each gender. However, challenges such as the heterogeneity of symptoms, especially in girls, the lack of knowledge about the early signs of ASD by health professionals and educators can be a barrier. CONCLUSION: AI tools showed high sensitivity, specificity, accuracy, and precision. However, it is crucial to overcome the challenges related to the heterogeneity of symptoms and limited access to specialized services so that AI benefits less favored populations, especially in developing countries.