文献速递泌尿生殖腹部MR

人工智能与影像组学在膀胱癌MRI中的应用、性能及临床转化障碍:范围综述

Artificial intelligence and radiomics in bladder cancer MRI: a scoping review of applications, performance, and barriers to clinical translation.
PubMed 文献 · Abdom Radiol (NY) · 2026-03-24

该范围综述纳入79项研究,梳理人工智能与影像组学在膀胱癌MRI中的应用。分期诊断性能在理想条件下AUC可达0.99,多中心稳健研究报道AUC为0.85-0.95,AI可辅助VI-RADS评估不确定病例。但证据以回顾性单中心设计为主、缺乏外部验证,报告透明度差且几乎无前瞻性效用试验,转化停滞风险高。

要点

  • 多中心稳健研究报道分期AUC为0.85-0.95,AI可能作为VI-RADS的补充,尤其用于不确定病例。
  • 证据支持通过简化方案(仅T2w)和工作流工具整合AI,但当前研究几乎均为回顾性单中心设计。
  • 报告透明度差、缺乏前瞻性效用试验,且研究集中于治疗前诊断,转化停滞风险高。

核心结论

  • 当前证据提示人工智能与影像组学在膀胱癌MRI中转化停滞风险高;实现其潜力需转向严格的多中心评估与透明度,并优先开展衡量临床效用的前瞻性研究。

研究设计与样本

  • 范围综述,纳入79项研究,领域以回顾性、单中心设计为主且缺乏外部验证。

诊断标准与关键数值

  • 理想条件下分期诊断AUC最高达0.99,稳健多中心研究报告AUC为0.85-0.95。
  • AI定位为VI-RADS的补充,尤其用于不确定病例。

临床应用与阅片要点

  • 证据支持通过简化方案(仅T2w)和工作流工具整合AI。
  • 优先事项包括强制遵循TRIPOD+AI报告标准、建立公共多中心基准、重新设计验证范式以优先开展衡量临床效用的前瞻性研究。

局限性

  • 领域以回顾性、单中心设计为主,缺乏外部验证。
  • 报告透明度差。
  • 几乎完全缺乏前瞻性效用试验。
  • 研究集中于治疗前诊断。

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

Artificial intelligence (AI) and radiomics show significant potential to augment bladder cancer (BC) MRI but face a critical translational gap. This scoping review of 79 studies maps a rapidly growing field dominated by retrospective, single-center designs lacking external validation. While diagnostic performance for staging is promising (AUC up to 0.99 in ideal settings), robust multi-center studies report more realistic AUCs of 0.85-0.95, positioning AI as a complement to VI-RADS, especially for equivocal cases. Evidence supports integration via simplified protocols (T2w-only) and workflow tools. However, systemic barriers impede progress: poor reporting transparency, a near-total absence of prospective utility trials, and a narrow focus on pre-treatment diagnosis. The current evidence suggests a high risk of translational stagnation. Realizing the transformative potential requires a fundamental reorientation toward rigorous, multi-center evaluation and transparency. Priorities must include mandatory adherence to reporting standards (TRIPOD + AI), the creation of public multi-center benchmarks, and a redesign of validation paradigms to prioritize prospective studies measuring clinical utility - such as impact on treatment decisions - over retrospective diagnostic accuracy alone. Only by prioritizing transparency, interdisciplinary collaboration, and patient-centered outcomes can AI evolve from a promising research instrument into a reliable clinical tool for BC management.

来源:Abdom Radiol (NY),查看原文。版权归原作者所有。

引用

  1. Salvaggio G, Comelli A, Albano D, et al. Artificial intelligence and radiomics in bladder cancer MRI: a scoping review of applications, performance, and barriers to clinical translation. Abdom Radiol (NY). 2026. PMID: 41874614. doi:10.1007/s00261-026-05457-6

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。