Abstract: Blood cell morphology assessment is a fundamental component of hematologic diagnosis but relies heavily on expert experience and is subject to inter-observer variability. With advances in deep learning，artificial intelligence has achieved substantial progress in digital hematopathology，ranging from automated single-cell recognition to whole-slide–level abnormality screening，which provides a new tool for improving diagnostic efficiency and consistency. However，clinical decision-making depends on the integrated interpretation of morphology，clinical context，and molecular data. This review summarizes the major research paradigms of AI-based blood cell analysis based on the current researches and clinical workflows，with a particular focus on the potential value of multimodal AI in abnormal cell detection，risk assessment，and prognostic clues. Current challenges regarding data scale，multicenter generalizability，interpretability，and clinical translation are also analyzed. With continued standardization efforts and multicenter collaboration，AI is expected to facilitate more precise and standardized diagnosis of blood cell abnormalities.

Key words: blood cell morphology assessment, artificial intelligence, multimodal analysis, clinical diagnosis assistance

摘要： 血细胞形态学检查是血液系统疾病诊断中的最基础环节，但高度依赖专业经验，易受主观因素影响。随着深度学习的发展，人工智能在血细胞数字图像分析领域取得显著进展，从单细胞自动识别到整张涂片级别的异常筛查，为提升诊断效率与一致性提供了新的工具。然而，临床诊断依赖形态、临床信息及分子检测的综合判断。文章结合现有研究与临床流程，系统梳理血细胞人工智能诊断的主要研究范式，重点探讨多模态人工智能在血细胞异常识别、风险评估及预后提示中的应用潜力，并分析其在数据规模、模型泛化、可解释性及临床转化方面的挑战。未来在标准化建设与多中心协作推动下，人工智能有望促进血细胞异常诊断向精准化与规范化发展。

关键词: 血细胞形态学检查, 人工智能, 多模态分析, 临床诊断辅助