Use of Artificial Intelligence in traumatic brain injury in children: scoping review

Abstract

Artificial Intelligence (AI) constitutes a valuable tool in clinical practice. Traumatic brain injury (TBI) in children represents one of the main causes of infant morbidity and mortality. The use of AI is expected to improve clinical outcomes in this population. Objective: To explore and analyze the existing literature on the use of AI in the management of Traumatic brain injury in the pediatric population. We searched PubMed/MEDLINE, PMC, Cochrane, Embase, Web of Science, IEEE Xplore, Scopus, Scielo and Lilacs databases. We included studies that applied Machine Learning (ML) models to predict diagnosis, treatment and prognosis, published between January 2015 and first semester 2024. A total of 1727 articles were identified, of which 31 were selected. The majority were published between 2021 and 2024, from the United States (51,6%) and Asian countries (29%). Supervised learning models, Random Forest and Support Vector Machine (SVM) were the most used (51,6%), followed by deep learning (32,2%), highlighting artificial neural networks (ANN). ML models were applied in diagnosis (64,5%) and prognosis (38,7%). In terms of performance, diagnostic models reported an AUC between 0,78-0,99 and ANN stood out (accuracy 99%, precision 100%); in prognosis, they reported an AUC of 0,71-0,99 and SVM stood out (accuracy 94%, precision 99%). There is an interest in the use of AI in the diagnosis and prognosis of pediatric TBI, highlighting deep learning models, which would outperform traditionally used clinical tools. © Revista de Neuro-Psiquiatría