文献速递神经康复

AI赋能医疗神经技术中的可解释性:范围综述

Explainability in AI-enabled medical neurotechnology: a scoping review.
PubMed 文献 · J Neuroeng Rehabil · 2026-02-04

该范围综述系统检索五个数据库,纳入161项关于AI赋能闭环神经技术(含神经刺激与脑机接口)的研究,发现仅14项(9%)采用显式可解释AI技术。文中识别出技术限制、解释质量与系统透明度、可解释性与用户信任三方面障碍,指出可解释性在实践中极少落地,制约临床可用性与信任。

要点

  • 仅9%(14/161)研究采用显式可解释AI技术。
  • 可解释性缺口涉及技术限制、透明度与用户信任三方面。
  • 可解释性不足制约闭环神经技术的临床可用性与问责。

核心结论

  • 闭环神经技术快速发展,但可解释性在实践中极少实现,制约透明度、问责与临床可用性。
  • 仅14项研究(9%)采用显式可解释AI技术,存在明显可解释性缺口。

研究设计与样本

  • 遵循JBI指导与PRISMA-ScR,系统检索五个数据库,纳入161项关于AI赋能医疗闭环神经技术(针对神经与精神疾病)的原始研究。
  • 采用描述性统计与定性内容分析识别可解释AI方法的存在与框架。

主要发现

  • 仅14项研究(9%)采用显式可解释AI技术。
  • 主题分析识别三类反复出现的障碍与焦点:技术限制与变通方案、解释质量与系统透明度挑战、可解释性与用户信任的关系。

临床怎么用

  • 在AI闭环神经技术研发与临床转化中应优先纳入显式可解释AI方法以提升透明度与问责。
  • 关注解释质量与用户信任的关系,将可解释性作为临床可用性的关键要求。

局限性

  • 作为范围综述,仅映射研究现状与概念化方式,未提供可解释AI对临床结局影响的定量证据。

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)。

原文

Artificial Intelligence (AI) approaches, including Machine Learning (ML), and other complex algorithms, are driving progress in medical closed-loop neurotechnology, including neurostimulation systems and brain–computer interfaces (BCIs). These advances are transforming the treatment landscape for neurological and psychiatric conditions. However, the inherent opacity of many AI models raises clinical, epistemological and ethical challenges. Explainability is widely recognized as a critical requirement for addressing these challenges, yet its concrete application in neurotechnology remains insufficiently explored. Objective. This scoping review maps how Explainable AI (XAI) methods are implemented in AI-enabled closed-loop neurotechnologies and examines how explainability is conceptualized and operationalized in this domain. Approach. Following JBI guidance and PRISMA-ScR, we systematically searched five databases for original research on AI-enabled medical closed-loop neurotechnologies targeting neurological and psychiatric conditions. A total of 161 studies were included and analyzed using descriptive statistics and qualitative content analysis to identify the presence and framing of XAI methods. Main results. Explainable AI adoption in medical neurotechnology is limited: only 14 studies (9%) employed explicit XAI techniques. Thematic analysis of the full corpus identified three recurring barriers and focal points: (A) technical constraints and proposed workarounds, (B) challenges in explanation quality and system transparency, and (C) the relationship between explainability and user trust. Significance. Although closed-loop neurotechnologies are rapidly advancing, explainability is rarely implemented in practice, constraining transparency, accountability, and clinical usability. Our findings reveal key factors behind the explainability gap and provides a framework to guide future research and development. Addressing this shortfall is essential for fostering ethically sound, clinically effective, and patient-trusted neurotechnological applications.

来源:J Neuroeng Rehabil,查看原文。版权归原作者所有。

引用

  1. Schopp L, Starke G, Ienca M, et al. Explainability in AI-enabled medical neurotechnology: a scoping review. J Neuroeng Rehabil. 2026. PMID: 41639858. doi:10.1186/s12984-026-01892-0

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。