病理图像标注作为人工智能（AI）模型训练关键环节，其准确性、一致性和标准化程度直接影响AI模型性能与临床应用价值。当前国内外缺乏统一完善的标注标准，导致标注数据差异大，限制数据共享复用及AI辅助病理诊断技术推广。系统梳理国内外相关标准与规范，分析病理图像标注的核心要素（标注对象、工具、评价指标）及其面临的挑战，如主观性差异、数据异质性、隐私安全与高成本等。国际上，国际医学数字成像和通信（DICOM）标准及相关项目为标注质量控制与数据复用提供了参考；国内有国家药品监督管理局（NMPA）发布的相关技术审查指导原则与《YY/T 1833.3-2022》标准，以及《人工智能面向机器学习的数据标注规程》等国家标准、山西省地方标准、中国通信标准化协会行业标准，同时专家共识也在推进专科数据集建设，在医疗器械监管、数据标注流程与质量评估等方面形成了初步标准体系。未来应加强AI辅助标注工具研发、专科标注规则制定与跨机构数据共享，构建开放、安全、协同的标注生态，推动数字病理发展和精准医疗实现，以充分发挥病理AI模型的临床价值。

As a critical step in training artificial intelligence (AI) models, the accuracy, consistency, and standardization of pathological image annotation directly determine the performance of AI models and their clinical applicability. Currently, the lack of comprehensive and unified annotation standards globally leads to significant variability in annotated data, which hinders data sharing, reuse, and the widespread adoption of AI-assisted pathological diagnostics. This paper systematically reviews existing standards and guidelines worldwide, analyzing core elements of pathological image annotation—such as annotation targets, tools, and evaluation metrics— along with challenges including subjective variability, data heterogeneity, privacy and security concerns, and high costs. Internationally, the Digital Imaging and Communications in Medicine (DICOM) standard and related initiatives provide references for quality control in annotation and data reuse. Domestically, guidelines such as those issued by the National Medical Products Administration (NMPA) for technical reviews, the standard “YY/T 1833.3-2022”, along with national standards like “Artificial Intelligence—Data Annotation Specification for Machine Learning”, regional standards from Shanxi Province, and industry standards from the China Communications Standards Association, have been established. Additionally, expert consensus is advancing the development of specialized datasets, forming a preliminary framework of standards in areas such as medical device regulation, data annotation workflows, and quality assessment. Moving forward, efforts should focus on enhancing the development of AI-assisted annotation tools, formulating specialized annotation rules, and promoting cross-institutional data sharing to build an open, secure, and collaborative annotation ecosystem. This will drive the progress of digital pathology and the realization of precision medicine, ultimately unlocking the full clinical potential of AI-based pathological models.

R285.5

药品监管科学全国重点实验室课题项目（2025SKLDRS0364）