公众号 / 文章技术与安全

平衡「黑箱」系统与可解释人工智能的必要性:放射学中必要的实施

The need for balancing 'black box' systems and explainable artificial intelligence: A necessary implementation in radiology.
PubMed 文献 · Eur J Radiol · 2025-02-26

放射学是受人工智能影响最显著的医学专科之一,但数据隐私、知情同意、外部干扰、偏倚、生成式人工智能的「幻觉」及责任归属等问题构成实施挑战。许多人工智能模型的「黑箱」性质限制了完全知情同意的可能性,作者主张在放射学中优先采用可解释人工智能(XAI),因其虽可能性能不及黑箱模型,但能增强透明度,使患者理解数据使用方式及人工智能对临床决策的影响,符合伦理标准。

要点

  • 人工智能系统作为支持工具而非替代人类专业知识和判断,医疗专业人员目前对涉及人工智能的医疗错误决策承担全部责任。
  • 许多人工智能模型的「黑箱」性质使输出背后的推理不透明,限制了完全知情同意的可能性。
  • 作者主张优先采用可解释人工智能(XAI),因其增强透明度,符合伦理标准。

核心结论

  • 作者主张在放射学中优先采用可解释人工智能(XAI),因其虽可能性能不及黑箱模型,但能增强透明度,使患者理解数据使用方式及人工智能对临床决策的影响,符合伦理标准。

临床应用与阅片要点

  • 人工智能系统作为支持工具而非替代人类专业知识和判断,医疗专业人员目前对涉及人工智能的医疗错误决策承担全部责任。
  • 作者主张优先采用可解释人工智能(XAI),因其增强透明度,使患者理解数据使用方式及人工智能对临床决策的影响,符合伦理标准。

局限性

  • 原文未报告局限性

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

Radiology is one of the medical specialties most significantly impacted by Artificial Intelligence (AI). AI systems, particularly those employing machine and deep learning, excel in processing large datasets and comparing images from similar contexts, fulfilling radiological demands. However, the implementation of AI in radiology presents notable challenges, including concerns about data privacy, informed consent, and the potential for external interferences affecting decision-making processes. Biases represent another critical issue, often stemming from unrepresentative datasets or inadequate system training, which can lead to distorted outcomes and exacerbate healthcare inequalities. Additionally, generative AI systems may produce 'hallucinations' arising from their reliance on probabilistic modeling without the ability to distinguish between true and false information. Such risks raise ethical and legal questions, especially when AI-induced errors harm patient health. Concerning liability for medical errors involving AI, healthcare professionals currently retain full accountability for their decisions. AI systems remain tools to support, not replace, human expertise and judgment. Nevertheless, the "black box" nature of many AI models - wherein the reasoning behind outputs remains opaque - limits the possibility of fully informed consent. We advocate for prioritizing Explainable Artificial Intelligence (XAI) in radiology. While potentially less performant than black-box models, XAI enhances transparency, allowing patients to understand how their data is used and how AI influences clinical decisions, aligning with ethical standards.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. De-Giorgio F, Benedetti B, Mancino M, et al. The need for balancing 'black box' systems and explainable artificial intelligence: A necessary implementation in radiology. Eur J Radiol. 2025. PMID: 40031377. doi:10.1016/j.ejrad.2025.112014

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。