数字病理与人工智能在髓系肿瘤诊断中的进展
综述指出,数字病理与人工智能已在外周血涂片、骨髓穿刺与活检、流式细胞术、细胞遗传学和二代测序等环节达到专家级性能,许多平台准确率超过90%,并能整合多模态输入支持分类、风险分层和治疗预测,但临床推广仍受成本、安全、监管和人群验证等挑战制约。
要点
- AI在外周血涂片、骨髓、流式、遗传学和测序中可达专家级性能。
- 多模态AI整合可支持分类、风险分层与治疗预测。
- 临床采用仍受成本、安全、监管和人群验证限制。
核心结论
- 数字病理与AI正在改变髓系肿瘤的诊断能力,并推动血液病理精准诊断。
- 早期监管成功案例表明真实世界实施具有可行性。
研究设计与样本
- 本文为综述,围绕数字病理与AI在髓系肿瘤诊断中的进展进行汇总。
- 涉及病种包括急性髓系白血病、骨髓增生异常综合征和骨髓增殖性肿瘤。
主要发现
- 现代AI系统在外周血涂片评估、骨髓穿刺与活检判读、流式细胞术、细胞遗传学和二代测序中达到专家级性能。
- 许多平台准确率超过90%,并可检测超出人类感知的细微形态学和免疫表型特征。
- AI工具越来越多地整合多模态输入以支持分类、风险分层和治疗预测,且常优于传统方法。
- 这些系统通过优化工作流程、提高一致性、减少观察者间变异和加快周转时间来辅助病理医生。
关键数值与推荐
- 许多AI平台准确率超过90%。
临床怎么用
- AI可辅助病理医生优化工作流程、提高诊断一致性并加快周转时间。
- AI多模态整合可支持髓系肿瘤的分类、风险分层和治疗预测。
- 临床采用需关注成本、安全、监管和跨人群验证。
- 更广泛推广需要开发者、实验室、病理医生和监管机构协同确保安全、透明和互操作性。
局限性
- 临床采用面临成本、安全、监管监督和跨多样人群验证等挑战。
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
Recent advances in digital pathology and artificial intelligence (AI) are transforming our ability to diagnose myeloid neoplasms, including acute myeloid leukemia (AML), myelodysplastic syndromes (MDS), and myeloproliferative neoplasms (MPN). Modern AI systems now achieve expert-level performance across peripheral blood smear assessment, bone marrow aspirate and biopsy interpretation, flow cytometry, cytogenetics, and next-generation sequencing; with many platforms surpassing 90% accuracy and enabling detection of subtle morphologic and immunophenotypic signatures beyond human perception. Moreover, AI tools increasingly integrate multimodal inputs to support classification, risk stratification, and treatment predictions, often outperforming conventional methods. These systems augment pathologists by streamlining workflows, enhancing consistency, reducing interobserver variability, and accelerating turnaround times, paving the way for advanced precision diagnostics in hematopathology. However, challenges remain in clinical adoption, including cost, security, regulatory oversight, and validation across diverse populations. Despite these hurdles, early regulatory successes such as digital peripheral blood smear morphology analyzers illustrate the feasibility of real-world implementation. Broader adoption will however require coordinated efforts among developers, laboratories, pathologists, and regulatory agencies to ensure safety, transparency, and interoperability.
引用
- Patel P, Pozdnyakova O. Advances in digital pathology and artificial intelligence in the diagnosis of myeloid neoplasms. Hum Pathol. 2026. PMID: 42214762. doi:10.1016/j.humpath.2026.106178
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。