文献速递肾移植

人工智能在肾移植病理中的应用:当前证据、局限性与Banff框架的相关性

Artificial Intelligence in Kidney Transplant Pathology: Current Evidence, Limitations, and Relevance to the Banff Framework.
PubMed 文献 · Kidney Int Rep · 2026-05-26

Banff分类自1991年起为肾移植病理提供全球标准,近年AI与深度学习加速了肾组织定量分析,尤其在IFTA和炎症方面减少观察者间变异。靶向算法如PPC和DL分割模型在识别肾分区和量化损伤上表现稳健,但受数据多样性、染色标准化和外验证限制。Banff数字病理工作组正将自动评估扩展至肾小球炎、管周毛细血管炎、动脉炎和肾小管炎等病变。

要点

  • AI/DL在肾移植病理中可定量IFTA和炎症,减少观察者间变异。
  • PPC和DL分割模型能稳健识别肾分区并量化损伤。
  • 血管病变和免疫细胞分析的AI方法有前景,但受数据多样性和外验证限制。

核心结论

  • AI与数字病理有望增强肾移植病理诊断精度并扩展Banff病变的自动评估。

研究设计与样本

  • 综述性文章,总结AI在肾移植病理中的当前证据、局限性及与Banff框架的相关性。

主要发现

  • AI和深度学习加速了肾组织定量分析,尤其针对IFTA和炎症,减少观察者间变异并支持预后评估。
  • PPC和DL分割模型在识别肾分区和量化损伤方面表现稳健。
  • 血管病变和免疫细胞谱的AI辅助方法有前景,但受数据多样性、染色标准化和外验证限制。
  • 肾单位大小、IFTA灶密度和系膜扩张等指标已被探索,计算指标和AI模型可能提高可重复性和临床相关性。

临床怎么用

  • AI工具可用于辅助肾移植活检的定量评估,但目前应视为增强人类专业知识的工具。

局限性

  • 数据多样性不足、染色标准化缺乏和外验证有限是当前AI方法的主要局限。

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)。

原文

The Banff Classification, established in 1991, provides a global standard for diagnosing and grading kidney transplant pathology, evolving through regular consensus meetings to incorporate new advances. Initially designed to address the lack of uniformity in renal allograft biopsy reporting, the Banff system now integrates semiquantitative scores for acute and chronic lesions. Recent progress in artificial intelligence (AI) and deep learning (DL) has accelerated quantitative analysis in kidney histology, particularly for interstitial fibrosis and tubular atrophy (IFTA) and inflammation, reducing interobserver variability and supporting prognostic assessment. Targeted algorithms such as positive pixel count (PPC) and DL-based segmentation models have demonstrated robust performance in identifying renal compartments and quantifying injury; and AI-assisted approaches for vascular lesions and immune cell profiling show promise but are limited in data diversity, stain standardization, and external validation. Additional metrics, such as nephron size, IFTA foci density, and mesangial expansion (ME), have been explored, with computational metrics and AI models showing potential for improved reproducibility and clinical relevance. Despite concerns about the implications of AI in pathology, optimism prevails regarding its ability to augment human expertise and enhance diagnostic precision. Ongoing work within the Banff Digital Pathology Working Group aims to extend automated assessment to additional Banff lesions, including glomerulitis, peritubular capillaritis, arteritis, and tubulitis. As these tools mature, they may enable more comprehensive quantitative characterization of transplant biopsies and facilitate exploration of histopathologic patterns that have been difficult to study. The Banff system's iterative evolution and embrace of AI-driven digital pathology underscore its enduring impact and promise in kidney transplant medicine.

来源:Kidney Int Rep,查看原文。版权归原作者所有。

引用

  1. Denic A, van Midden D, Boor P, et al. Artificial Intelligence in Kidney Transplant Pathology: Current Evidence, Limitations, and Relevance to the Banff Framework. Kidney Int Rep. 2026. PMID: 42437163. doi:10.1016/j.ekir.2026.106614

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。