全自动数字图像分析在外周血白细胞形态学复检中的临床应用

Clinical application of automated digital image analysis for morphology review of peripheral blood leukocyte

目的探讨外周血血细胞分析触发白细胞复检规则时，全自动数字图像分析系统在白细胞形态学复检中的应用。 方法选择北京大学第一医院304份健康体检者的血液标本，建立全自动血细胞形态学数字图像分析仪的白细胞分类计数参考区间。随机选择2013年11月至2014年4月北京大学第一医院697份临床血液标本，在全自动血细胞分析仪上完成全血细胞计数（CBC）并自动推制血涂片和瑞氏染色，然后将血涂片置于全自动血细胞形态学数字图像分析仪检测，并且经有形态学经验的工作人员进行审核（再分类）。将同一张血涂片在人工显微镜下完成白细胞分类计数。以人工显微镜分类计数结果为＂金标准＂，计算全自动血细胞形态学数字图像分析仪对各种白细胞分类计数或形态异常的诊断效率，包括敏感度、特异度、精确度等指标。选择30份临床诊断明确的血液病和感染患者标本，统计上述两种方法对常见异常白细胞的识别是否存在差异。 结果全自动血细胞形态学数字图像分析仪分类计数各种白细胞或形态异常检测的特异度较高，除单核细胞外均〉90%；中性粒细胞中毒性形态异常（包括杜勒小体、中毒颗粒和空泡变性）检测的敏感度可达到100%；其次是原始细胞（91.7%）、不成熟粒细胞（60%～81.5%）、异型淋巴细胞（61.5%）；分类计数5种白细胞数量异常的敏感度分别为：中性粒细胞91.8%、淋巴细胞88.5%、单核细胞69.1%、嗜酸粒细胞78.9%、嗜碱粒细胞36.3%。人工镜检对3种常见异常白细胞（原始细胞、不成熟粒细胞和异型淋巴细胞）的阳性识别率分别为46.7%、53.3%和10%，全自动数字图像分析仪的阳性识别率分别为43.3%、60%和10%，差异均无统计学意义（χ2＝0.067、0.271、0.000，均P〉0.05）。可以将全自动血细胞数字图像分析用于白细胞异常时的血涂片形态学检查，对血细胞分析仪复检规则进行优化。 结论全自动血细胞形态学数字图像分析对外周血异常白细胞的分类计数和形态学异常检测具有较高的敏感度和特异度，可用于触发血细胞分析仪复检规则时异常白细胞的形态学复检。

ObjectiveTo explore the clinical application of automated digital image analysis in leukocyte morphology examination when review criteria of hematology analyzer are triggered. MethodsThe reference range of leukocyte differentiation by automated digital image analysis was established by analyzing 304 healthy blood samples from Peking University First Hospital. Six hundred and ninty-seven blood samples from Peking University First Hospital were randomly collected from November 2013 to April 2014, complete blood cells were counted on hematology analyzer, blood smears were made and stained at the same time. Blood smears were detected by automated digital image analyzer and the results were checked （reclassification） by a staff with abundant morphology experience. The same smear was examined manually by microscope. The results by manual microscopic differentiation were used as＂golden standard＂, and diagnostic efficiency of abnormal specimens by automated digital image analysis was calculated, including sensitivity, specificity and accuracy. The difference of abnormal leukocytes detected by two different methods was analyzed in 30 samples of hematological and infectious diseases. ResultsSpecificity of identifying abnormalities of white blood cells by automated digital image analysis was more than 90% except monocyte. Sensitivity of neutrophil toxic abnormities （including D？hle body, toxic granulate and vacuolization） was 100%; sensitivity of blast cells, immature granulates and atypical lymphocytes were 91.7%, 60% to 81.5% and 61.5%, respectively. Sensitivity of leukocyte differential count was 91.8% for neutrophils, 88.5% for lymphocytes, 69.1% for monocytes, 78.9% for eosinophils and 36.3 for basophils. The positive rate of recognizing abnormal cells （blast, immature granulocyte and atypical lymphocyte） by manual microscopic method was 46.7%, 53.3% and 10%, respectively. The positive rate of automated digital image analysis was 43.3%, 60% and 10%, respectively. There was no statistic significance between two methods （χ2 = 0.067, 0.271, 0.000, all P〉0.05）. Automated digital image analysis could be used to morphology examination with abnormal leukocytes and optimize review criteria of hematology analyzer. ConclusionSensitivity and specificity of recognizing abnormal blood leukocytes by automated digital image analysis are satisfactory, which can be used as a tool of leukocyte morphology review when review criteria of hematology analyzer are triggered.