Avaliação da efetividade da inteligência artificial no diagnóstico imediato de patologias torácicas em radiografias de urgência: revisão integrativa da literatura

Abstract

Considering the problem of delayed interpretation of chest radiographs in emergency settings, especially due to the shortage of radiologists, this study is justified by the need for tools capable of optimizing diagnosis and improving clinical decision-making. The objective is to evaluate the effectiveness of Artificial Intelligence (AI), particularly deep learning algorithms, in the immediate diagnosis of thoracic pathologies on emergency chest radiographs. To this end, an integrative literature review was conducted between August and September 2025 using the Google Scholar, PubMed, LILACS, and SciELO databases, including studies published in the last five years that address the use of AI in detecting pneumothorax, heart failure, pneumonia, and tuberculosis. Thus, it was observed that deep learning models achieve average sensitivity and specificity above 90%, with performance comparable to that of experienced radiologists and a significant reduction in triage time, reinforcing their clinical potential as diagnostic support tools. This allows us to conclude that the application of AI in chest radiography represents a relevant advance for emergency radiology, contributing to faster, more accurate, and accessible diagnoses, although methodological standardization and multicenter validation are still required to ensure its safe integration into medical practice.