人工智能辅助多参数流式细胞术检测淋巴细胞亚群免疫表型的应用研究  

Application of multiparameter flow cytometry in detecting immunophenotyping of lymphocyte subsets by artificial intelligence assistance

作　　者：贾晓冬 汲珊珊 刘蕊[2] 楚玉兰 JIA Xiaodong;JI Shanshan;LIU Rui;CHU Yulan(Graduate School of Tianjin Medical University,Tianjin 300070,China;Tianjin Union Medical Center,Tianjin Medical University,Tianjin 300121,China;Tianjin Kingmed Medical Laboratory Co.,Ltd.,Tianjin 300392,China)

机构地区：[1]天津医科大学研究生院,300070 [2]天津医科大学人民医院临床学院,300121 [3]天津金域医学检验实验室有限公司,300392

出　　处：《重庆医学》2022年第24期4294-4299,4303,共7页Chongqing medicine

基　　金：中央引导地方科技发展专项区域创新体系建设(百城百园)项目(20ZYQCSY00010)。

摘　　要：目的通过建立人工智能(AI)模型,评价AI辅助多参数流式细胞术(MFC)检测人外周血中淋巴细胞亚群(LS)免疫表型方法的可行性。方法运用流式细胞术对2020年6-7月天津金域医学检验实验室收集的1263例患者外周血样本T淋巴细胞(CD3^(+)、CD4^(+)、CD8)、B淋巴细胞(CD3^(-)CD19^(+))及NK细胞(CD3-CD16^(+)CD56^(+))进行检测,基于高斯混合模型的多维数据聚类算法,应用聚类算法和核密度估计方法对淋巴细胞免疫表型自动判定和统计分析,评价两种检测方法的一致性。结果经与人工分析结果比较,1263例人工分析数据中有1199例通过AI分析一致性检测的通过比例为94.93%。诊断结果差异项经高年资医师复核后与AI诊断结果一致。AI单个样本平均分析时间为(1.36±0.25)s,较传统人工分析提高约50倍以上。且两种方法具有良好的一致性。AI分析方法质控的重复系数为2.8331%,95%置信区间:2.7268%~2.9481%,均在临床可接受临界值(±5%)范围内,表明AI分析可快速检测出人外周血中淋巴细胞各亚群百分比,且重复性好。结论初步建立的AI辅助MFC检测LS免疫表型模型重复性好,分析速度快,准确率高,已基本满足了临床使用需求,可用于基于6色抗体组合方案辅助MFC检测人外周血LS百分比的检测。Objective To evaluate the feasibility of AI-assisted multi-parameter flow cytometry(MFC)for detecting the immunophenotype of lymphocyte subsets(LS)in human peripheral blood by establishing the artificial intelligence(AI)model.Methods MFC was used to detect T lymphocytes(CD3^(+),CD4^(+),CD8^(+)),B lymphocytes(CD3-CD19^(+))and NK cells(CD3^(-)CD16^(+)CD56^(+))in 1263 peripheral blood samples collected by Tianjin Kingmed Medical Laboratory.The multi-dimensional data clustering algorithm based on Gaussian mixture model was used to automatically determine and statistically analyze the lymphocyte immunophenotype by clustering algorithm and kernel density estimation method.The consistency between the two methods was evaluated.Results In the comparison with the manual analytical results,among 1263 manual analysis data,1199 cases passed the consistency detection of AI analysis,with a passing rate of 94.93%.The difference items in the diagnosis results were consistent with the AI diagnosis after reexamination by senior physicians.The average analysis time of AI single sample was(1.36±0.25)s,which was more than 50 times faster than that of the traditional manual analysis.The two methods showed a good consistency.The quality control repetition coefficient(CR)of AI analysis method was 2.8331%,95%confidence interval was 2.7268%to 2.9481%,which all were within the clinically acceptable critical value(±5%),indicating the the AI analysis could rapidly detect the LS percentage of peripheral blood LS and peripheral blood lymphocytes subsets percentage with good repeatability.Conclusion The preliminarily established AI-assisted MFC model for detecting LS has good repeatability with rapid analysis speed and high accuracy,which basically meets the clinical application demand and could be used to detect LS percentage of human peripheral blood based on 6-color antibody combination scheme assisted MFC.

关 键 词：人工智能 流式细胞术 淋巴细胞亚群 

分 类 号：R331[医药卫生—人体生理学]