<![CDATA[Background: Artificial Intelligence (AI) has fundamentally transformed the paradigm of radiological image interpretation. Deep learning capabilities offer opportunities for improved diagnostic accuracy, reduced reading time, and minimized inter-reader variability. However, comprehensive evidence regarding AI’s impact across imaging modalities still requires systematic synthesis, particularly concerning which readers benefit most and under what conditions. This review aimed to synthesize scientific evidence on AI’s impact on diagnostic accuracy, interpretation time, diagnostic confidence, and the factors shaping human–AI interaction in radiology interpretation. Methods: A systematic review was conducted following the PRISMA 2020 guidelines. Searches were performed on PubMed/MEDLINE, Scopus, Google Scholar, and IEEE Xplore for studies published between 2021 and 2025. Of 847 identified records, 8 studies met the inclusion criteria as controlled empirical studies, comprising multi-reader, prospective observational, comparative retrospective, and experimental designs. Methodological quality and risk of bias were appraised using the QUADAS-2 tool. Results: The eight included studies covered mammography, chest radiography, computed tomography, magnetic resonance imaging, angiography, and musculoskeletal radiography, encompassing more than 100,000 image readings. AI consistently improved diagnostic sensitivity, with increases ranging from 11.4% to 23.0% across studies, shortened reading time for normal cases, and enhanced diagnostic confidence among readers. Junior radiologists and non-radiologist clinicians consistently benefited the most, suggesting AI functions as a diagnostic equalizer. However, the same group also showed greater susceptibility to automation bias, and AI performance was conditional on model accuracy, explanation design, and technical reliability across subpopulations. Conclusions: AI consistently improves radiological diagnostic performance without compromising reader productivity, with the magnitude of benefit conditional on model accuracy, explanation design, and reader expertise. Implementation in the Indonesian health system should prioritize local validation, AI literacy training for health workers, and integration into JKN/BPJS-based teleradiology models to address the chronic shortage and maldistribution of radiologists. Keywords: Artificial Intelligence; Deep Learning; Diagnostic Accuracy; Medical Image Interpretation; Radiology]]>
Copyrights © 2026
