文献速递技术与安全

影响放射学中人工智能接受度的因素:一项跨放射学工作流程的系统综述

Factors influencing AI acceptance in radiology: a systematic review across the radiology workflow.
PubMed 文献 · Eur Radiol · 2025-11-21

该系统综述纳入37项研究,识别出22个影响临床AI工具接受度的因素并归为6大主题,发现临床准确性是接受的前提条件但不足以推动实施,对放射科医生核心胜任力的全自动化任务阻力最强,而支持性和低风险应用更易被接受。

要点

  • 临床准确性是AI接受的前提条件,但仅凭准确性不足以提升实施效果。
  • 对图像解读这类放射科医生核心胜任力的全自动化任务,阻力最强。
  • 支持性和低风险AI应用更易被接受,AI素养、透明度、监管和伦理问题仍是挑战。

核心结论

  • 综述识别出22个接受度因素并归为6大主题,临床准确性是接受的条件性因素但不足以推动实施。
  • 对放射科医生核心胜任力的全自动化任务阻力最强,支持性和低风险应用更易被接受。
  • 存在两大证据缺口:非临床利益相关者代表性不足,图像解读之外的研究有限。
  • 需要涉及多元利益相关者的个案评估来支持AI在放射学中的负责任实施。

研究设计与样本

  • 系统综述,检索Ovid Medline、Embase和Web of Science Core,使用机器学习工具ASReview优先筛选相关记录。
  • 纳入标准针对多元利益相关者及评估临床AI模型(而非AI一般概念)接受度的研究。
  • 共纳入37项研究,提取的研究特征包括AI模型的预期用途、放射学亚专科、影像方式、利益相关者组和技术准备度。
  • 按放射学工作流程各阶段进行叙述性综合。

诊断标准与关键数值

  • 多数研究调查图像解读的AI应用,主要利益相关者组为放射科医生和患者。
  • 识别出22个接受度因素,归为6大主题。
  • 临床准确性是接受的条件性因素,但在AI素养、监管与伦理关切、用户指导有限、缺乏透明度和训练数据不具代表性方面仍存在挑战。

临床应用与阅片要点

  • 为在放射学中有效实施AI,算法透明度、数据代表性、AI素养和监管是必不可少的。
  • 多元利益相关者的参与对于从AI接受推进到有意义的临床整合是必不可少的。

局限性

  • 非临床利益相关者代表性不足。
  • 图像解读之外的研究有限。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

OBJECTIVES: Despite AI's promise for radiology, clinical implementation remains limited. AI acceptance is a key factor in bridging the gap between technical validation and adoption. This systematic review identifies factors influencing the acceptance and use of clinical AI tools across the radiology workflow.

MATERIALS AND METHODS: A systematic search of Ovid Medline, Embase, and Web of Science Core was conducted using ASReview, a machine learning tool that prioritises relevant records for selection. Inclusion criteria targeted diverse stakeholders and studies that evaluate the acceptance of clinical AI models, rather than AI as a general concept. Extracted study characteristics included the AI model's intended purpose, subspecialty of radiology, imaging modality, stakeholder group, and technology readiness level. A narrative synthesis was performed according to the stages of the radiology workflow.

RESULTS: Thirty-seven studies were included. Most studies investigated AI applications for image interpretation, with radiologists and patients as primary stakeholder groups. Twenty-two acceptance factors were identified and grouped into 6 overarching themes. Indicating that clinical accuracy is conditional for acceptance, however, challenges remain in AI literacy, addressing regulation and ethical concerns, limited user guidance, lacking transparency and unrepresentative training data. Resistance was strongest against fully automated tasks central to radiologists' core competencies, while supportive and lower-risk applications were more readily accepted.

CONCLUSION: This review highlights two major evidence gaps: the underrepresentation of non-clinical stakeholders and limited studies outside image interpretation. Case-specific evaluations involving diverse stakeholders are needed to support the responsible implementation of AI in radiology.

KEY POINTS: Question How does a broad range of stakeholders perceive AI acceptance along the radiology workflow? Findings AI acceptance is multidimensional, while clinical accuracy is essential, it alone is insufficient to increase implementation. Strongest resistance occurs toward automated image interpretation tasks. Clinical relevance To enable effective implementation of AI in radiology, algorithmic transparency, data representation, AI literacy, and regulation are essential. Engagement of diverse stakeholders is essential to advance from AI acceptance toward meaningful clinical integration.

来源:Eur Radiol,查看原文。版权归原作者所有。

引用

  1. Verwey J, Zwart B, IJzerman M, et al. Factors influencing AI acceptance in radiology: a systematic review across the radiology workflow. Eur Radiol. 2025. PMID: 41269301. doi:10.1007/s00330-025-12129-4

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。