论文部分内容阅读
为了实现血液中肝癌细胞的自动识别,本文基于主成分分析(PCA)和反向传播(BP)神经网络算法对三种细胞(小鼠的白细胞、红细胞和人体肝癌细胞Hep G2)进行了识别研究。利用光纤共聚焦后向散射(FCBS)光谱仪获取光谱数据后进行PCA,选取前两个主成分作为光谱的特征,建立一个具有2个输入层节点、11个隐层节点、3个输出节点的神经网络模式识别模型。选取195例对象数据训练该模型,随机抽取150组数据作为训练集,45组数据作为测试集,验证模型给出的细胞是否识别准确。结果显示三种细胞的整体识别准确率在90%以上,平均相对偏差只有4.36%。实验结果预示采用PCA+BP算法能够从红细胞和白细胞中自动识别肝癌细胞,这将为研究肝癌的转移与肝癌的生物代谢特性提供有利的工具。
In order to realize the automatic identification of hepatoma cells in blood, we identified the three kinds of cells (mouse white blood cells, red blood cells and human liver cancer cells Hep G2) based on Principal Component Analysis (PCA) and Back Propagation (BP) neural network algorithms . After obtaining the spectral data by optical fiber confocal backscatter (FCBS) spectrometer and PCA, the first two principal components are selected as the characteristics of the spectrum to establish a neural network with two input nodes, 11 hidden nodes and three output nodes Network pattern recognition model. Select 195 object data to train the model, randomly select 150 sets of data as a training set, 45 sets of data as a test set to verify whether the model is accurate identification of cells. The results showed that the overall recognition accuracy of the three kinds of cells was over 90%, and the average relative deviation was only 4.36%. The experimental results indicate that the PCA + BP algorithm can automatically identify hepatoma cells from erythrocytes and leukocytes, which will provide a useful tool for studying the metastasis of hepatocellular carcinoma and the biological metabolism characteristics of hepatocellular carcinoma.