摘要
目的探究人工神经网络(ANN)应用于流式细胞术(FCM)数据自动分析的可行性。方法以10名健康志愿者骨髓标本的FCM数据作为数据集,在Python中构建ANN模型,模型在训练中自学习提取特征,得到最优参数。以人工设门分析结果作为金标准,采用交叉验证的方式评价ANN模型在细胞亚群分群上的效果,并与决策树和K-means模型进行结果比较。结果ANN模型在数据集上拟合良好,模型在散点图上的分群轮廓与人工设门基本一致,能够很好的复现人工分析结果。ANN和决策树模型的分群效果优于K-means模型,分群准确率分别为0.970、0.972和0.899。结论ANN在FCM数据自动分析中具有一定的应用价值,可以为后续研究提供参考。
Objective To explore the feasibility of artificial neural network(ANN)for automatic analysis of flow cytometry(FCM)data.Methods The FCM data of bone marrow samples of 10 healthy volunteers were used as a dataset to construct an ANN model in Python,and the model extracted features by self-learning during training to obtain the optimal parameters.The results of manual gate analysis were used as the gold standard,and the effect of ANN model on cell subpopulation was evaluated by cross-validation,and the results were compared with decision trees and K-means models.Results The ANN model fits well on the dataset,and the grouping profile of the model on the scatter plot was basically consistent with the manual gate,which could reproduce the manual analysis results well.The grouping effect of ANN and decision tree model was better than that of K-means model,and the grouping accuracy was 0.970,0.972 and 0.899,respectively.Conclusion ANN has certain application value in the automatic analysis of FCM data,which can provide reference for subsequent research.
作者
雷伟
李智伟
郭玉娟
摆文丽
芮东升
王奎
LEI Wei;LI Zhi-wei;GUO Yu-juan;BAI Wen-li;RUI Dong-sheng;WANG Kui(Shihezi University School of Medicine,Shihezi 832000,Xinjiang,China;Clinical Testing Center of Xinjiang Uygur Autonomous Region People's Hospital,Urumqi 830001,Xinjiang,China)
出处
《医学信息》
2023年第18期74-77,98,共5页
Journal of Medical Information
基金
国家自然科学基金项目(编号:81860374)。
关键词
流式细胞术
细胞分群
人工神经网络
自动分析
Flow cytometry
Cell clustering
Artificial neural networks
Automated analysis
