摘要
目的探讨MLP神经网络在宫颈细胞图像识别中的应用。方法将测量的子宫颈细胞和细胞核的27个特征量作为MLP神经网络的输入参数,利用软件STATISTICA7.0建立网络模型,使用四种不同的算法训练网络,对700个子宫颈细胞进行分类识别,使用VC++.NET语言模拟调用网络。结果在四种算法中,使用共轭梯度法训练的MLP神经网络学习63次后,训练集识别率为98.67%,测试集识别率达到94.44%。不同算法的MLP神经网络的输入参数的敏感度排序均与细胞病理学特征基本一致。结论使用共轭梯度法训练的MLP神经网络可以较好地对宫颈细胞特征进行分类识别,在计算机辅助诊断方面具有广阔的应用前景。
Objective To explore the application of MLP neural network in image diagnoses of cervix cells. Methods Twenty - seven characteristic parameters of cervices cells and karyon were measured and used as input parameters in the MLP neural network. STATISTICA 7.0 was used to build up the network model, four methods were used to train the network model, seven hundred cervix cells were diagnosed in the network model, and VC ＋ ＋ . NET was used to call the network model. Resuits After MLP neural network was trained 63 times by conjugate gradient method selected among the four methods, diagnosis accuracy of training set was 98.67 % and diagnosis accuracy of testing set was 94.44 % Sensitivities order of input parameters in the MLP neural network of different methods was approximately in accordance with characters of cell pathology. Conclusion Cervix cells can be classified and diagnosed well by MLP neural network which is trained by conjugate gradient method, MLP neural network shows its great significance in computer aided diagnosis.
出处
《中国卫生统计》
CSCD
北大核心
2006年第4期293-296,共4页
Chinese Journal of Health Statistics
基金
辽宁省教育厅科研资助项目(202013137
05L534)
