Clinical Applications of AI for Chest Radiographs and CT: A Systematic Review of Diagnostic Performance and Workflow Impact

Abstract

Artificial intelligence (AI) is increasingly integrated into medical imaging to support rapid diagnosis, particularly in emergency care and pandemic contexts, while also alleviating radiologists’ workload and improving workflow efficiency. While promising, its generalizability and susceptibility to bias remain concerning. A systematic review was conducted following PRISMA guidelines. Thirty-six studies assessing AI in medical imaging were included.  The findings highlight both the promise and constraints of AI in medical imaging, with implications for future integration into radiological practice. AI achieved high diagnostic performance, with reported sensitivity and specificity exceeding 90% in differentiating COVID-19 pneumonia and other respiratory diseases. Applications also improved workflow efficiency and supported radiologist decision-making. However, performance was reduced when models were trained on limited or non-diverse datasets, leading to potential diagnostic errors. AI has substantial potential to enhance diagnostic accuracy and efficiency in chest imaging. Addressing dataset diversity and algorithmic bias is essential for safe and reliable clinical integration.