Objective: This study synthesizes current evidence regarding the performance and clinical applications of artificial intelligence (AI)-assisted systems in radiology report generation. The review compares AI-generated and human-authored reports across accuracy, clarity, and efficiency, evaluates workflow integration, and identifies key evidence gaps.
Methods: A synthesis of recent literature (2021–2025) was conducted, focusing on original research and systematic reviews involving multimodal AI frameworks and large-scale language models applied to imaging modalities including X-ray, CT, and MRI.
Results: Evidence indicates that AI has evolved from simple image captioning to multimodal, clinically aware frameworks capable of advanced reasoning. In specialized domains such as neurological and hepatobiliary imaging, AI achieves diagnostic accuracy rates between 94% and 99%. Automated systems generate reports in under one second, compared to 1.5–2 minutes for manual drafting, and can reduce radiologist review time by up to 58%. AI-generated reports also improve patient comprehension, with scores nearly doubling (2.71 to 4.69/5) when translating technical language. However, human-authored reports remain superior in clinical depth and nuanced reasoning. Feasibility and Workflow Integration: Integration is increasingly feasible through explainable hybrid models using interpretable features and lesion-aware logic, supporting human-AI collaboration.
Gaps and Future Research:: Barriers include hallucinations, limited longitudinal judgment, and inadequate evaluation metrics. Future work should focus on robust, clinically aligned evaluation methods and models capable of assessing temporal patient data.
Conclusion:: AI is a valuable decision-support tool that enhances productivity and patient engagement, but expert human oversight remains essential for clinical safety.
Methods: A synthesis of recent literature (2021–2025) was conducted, focusing on original research and systematic reviews involving multimodal AI frameworks and large-scale language models applied to imaging modalities including X-ray, CT, and MRI.
Results: Evidence indicates that AI has evolved from simple image captioning to multimodal, clinically aware frameworks capable of advanced reasoning. In specialized domains such as neurological and hepatobiliary imaging, AI achieves diagnostic accuracy rates between 94% and 99%. Automated systems generate reports in under one second, compared to 1.5–2 minutes for manual drafting, and can reduce radiologist review time by up to 58%. AI-generated reports also improve patient comprehension, with scores nearly doubling (2.71 to 4.69/5) when translating technical language. However, human-authored reports remain superior in clinical depth and nuanced reasoning. Feasibility and Workflow Integration: Integration is increasingly feasible through explainable hybrid models using interpretable features and lesion-aware logic, supporting human-AI collaboration.
Gaps and Future Research:: Barriers include hallucinations, limited longitudinal judgment, and inadequate evaluation metrics. Future work should focus on robust, clinically aligned evaluation methods and models capable of assessing temporal patient data.
Conclusion:: AI is a valuable decision-support tool that enhances productivity and patient engagement, but expert human oversight remains essential for clinical safety.