LIBSVM在急性髓细胞白血病流式细胞术残留病灶分析中的应用

Application of LIBSVM in analysis of minimal residual disease of acute myeloid leukemia

目的探讨利用支持向量机软件LIBSVM辅助流式细胞术,实现急性髓细胞白血病微小残留病灶(MRD)的自动化、客观化、标准化。方法从2010至2012年的流式数据库中选取36名急性髓细胞白血病患者,共计159例次MRD检测结果,分别导出其初发免疫分型和各次MRD复查的数据。以健康人骨髓数据和患者初发免疫分型数据为训练对象,利用LIBSVM优化参数并建立其对应的特异性预测模型,并与手工分析的MRD数据进行统计学比较。结果LIBSVM建立并优化了患者个体特异性的预测模型,基于此模型的MRD自动化分析结果与经验丰富的流式专业人员手工分析结果相似,相关性强。结论 LIBSVM结合流式细胞术有助于实现MRD分析的自动化、客观化和标准化。

Objective To investigate establishing automated, objective, and standard flow eytometry analysis method of minimal residual disease （MRD） in acute myeloid leukemia with LIBSVM. Methods Totally 159 MRD detection results from 36 cases with acute myeloid leukemia from 2010 to 2012 were enrolled in the study. All the onset immunophenotyping data and minimal residual disease data were exported as FCS files. Using training data from leukemic cells of patients and normal bone marrow data of healthy donors, patients specific SVM prediction model was trained, optimized and established by LIBSVM. Results The SVM model was established and optimized for independent patients. Strong correction was found between these automated analysis results and manual analysis ones. Conclusion LIBSVM can be applied for automated, objective and standard analysis of flow cytometry data.