流式细胞术结合支持向量机鉴别反应性和慢性粒细胞白血病肿瘤性粒细胞增多

目的探讨流式细胞术结合支持向量机（SVM）能否作为鉴别反应性粒细胞增多和慢性粒细胞白血病（CML）的肿瘤性粒细胞增多的新方法。方法用流式细胞术检测9例CML患者和9例健康人骨髓粒细胞的CD65s、CD15、CD11b、CD45表达，导出数据文件。利用SVM学习多维数据，建立并优化预测模型。基于该模型，对67例中性粒细胞增多患者进行分类预测，并通过受试者操作特征曲线（ROC）分析确定截断值，计算特异度和敏感度。结果ROC分析显示截断值为51．79％，以此截断值可有效区分反应性粒细胞增多和CML肿瘤性粒细胞增多（曲线下面积为0．97），预测的特异度95．80％，敏感度95．30％。结论SVM可通过学习多维流式数据，辅助流式细胞术鉴别反应性粒细胞增多和CML肿瘤性粒细胞增多。

Discriminating chronic myeloid leukemia from reactive granulocytosis by flow cytometry combining with support vector machines

Objective To investigate whether flow cytometry combined with support vector machines （SVM） algorithm could be a novel approach to differentiate neoplastic granuloeytosis of chronic myeloid leukemia （CML） and reactive granuloeytosis. Methods Expressions of CD65s, CD15, CD11b and CD45 in 9 patients with CML and 9 healthy volunteers were detected by flow cytometry, then flow cytometry standard data file were created. The multi-dimensional data were analyzed by SVM, and an optimized model was established. Based on the trained SVM model file and test data, classification of 67 patients presented with granulocytosis were predicted. The cut-off value was determined by receiver operating characteristic analysis, and the specificity and sensitivity were also calculated. Results The cut-off value used to classify the reactive granulocytosis and neoplastic granulocytosis of CML was 51.79%, while the specificity and sensitivity were 95.80%, 95.30%, respectively. Conclusion SVM may be helpful in differential diagnosis of reactive granulocytosis and neoplastic granulocytosis of CML.