Seah Jarrel C Y, Tang Jennifer S N, Tran Aengus
Division of Neuroradiology, Massachusetts General Hospital, Boston, Mass.
Department of Radiology, Alfred Health, Melbourne, Victoria, Australia.
Radiology. 2025 Jul;316(1):e243378. doi: 10.1148/radiol.243378.
Radiology faces a global shortage of radiologists. This shortage, particularly in the United States, affects timely diagnosis and exacerbates burnout due to escalating workloads. Because training new radiologists and developing the workforce takes a substantial amount of time, there is growing interest in using artificial intelligence (AI) as a key solution to help meet demand and improve efficiency. Recent years have seen the development of , where a single algorithm is trained with multitask learning to classify and detect multiple abnormalities on images from high-volume modalities such as radiography and CT. Further, the parallel and rapid development of capability in large language models (LLMs) intersects with comprehensive AI, enabling the production of accurate and human-like draft reports. Multimodal LLMs are capable of accepting images as well as text as inputs, giving them the ability to generate a draft report directly from medical images; this marks a major step forward in AI application in radiology. These AI-produced reports can dramatically enhance efficiency, scaling the net capacity of the current workforce. The authors envision a future in which AI is used to fully automate reporting for high-volume modalities in routine clinical settings, and they discuss evaluation frameworks for safe and effective integration of these AI report generation tools into clinical workflows.
放射学领域面临着全球范围内放射科医生短缺的问题。这种短缺，尤其是在美国，影响了及时诊断，并由于工作量不断增加而加剧了职业倦怠。由于培养新的放射科医生和发展劳动力需要大量时间，人们越来越有兴趣将人工智能（AI）作为关键解决方案，以帮助满足需求并提高效率。近年来，已经出现了这样的发展情况，即通过多任务学习训练单一算法，以对来自X射线摄影和CT等高流量模态的图像上的多种异常进行分类和检测。此外，大语言模型（LLMs）能力的并行快速发展与综合AI相交，使得能够生成准确且类似人类的报告草稿。多模态LLMs能够接受图像和文本作为输入，使其能够直接从医学图像生成报告草稿；这标志着AI在放射学应用方面向前迈出了重要一步。这些由AI生成的报告可以显著提高效率，扩大当前劳动力的净容量。作者设想了一个未来，即AI用于在常规临床环境中对高流量模态的报告进行完全自动化，并且他们讨论了将这些AI报告生成工具安全有效地整合到临床工作流程中的评估框架。
