基于遗传算法优化的BP神经网络大肠癌证型分类

Classification of large intestine carcinoma syndromes based on genetic algorithm-optimized back propagation neural network

针对医学领域传统BP神经网络应用于诊断辅助构建模型过程存在的疾病输入特征维数繁多而导致网络训练时间长、效率低、诊断模型泛化能力弱等问题,提出了将基于遗传算法降维优化的BP神经网络(GABP)诊断模型用于大肠癌虚实证型的分类研究。利用遗传优化算法从大肠癌的28项体征输入指标中筛选出的10个指标作为GABP诊断模型的输入,并与传统的未经优化的BP诊断模型进行对比,发现优化后的神经网络模型建模所需时间由5.1844秒缩短至0.1976秒,虚证正判率由76.4567%提升至89.1809%,实证正判率由64.8441%提升至70.1170%。仿真结果表明,基于GABP的神经网络泛化能力更好,分类效果更佳,为大肠癌中医证型的临床判别提供了很好的计算机网络模型。

A diagnosis model of genetic algorithm（ GA） dimension reduction-optimized back propagation（ BP）neural network was established since the long time,low efficiency and poor capability of diagnosis model due to the large number of disease input feature dimensions when the traditional BP neural network was used in establishing diagnosis-aided model,which has been used in classification of large intestine carcinoma syndromes. Ten input feature indications,screened from the 28 input feature indications of large intestine carcinoma,were used as the input of GABP diagnosis model. The optimized GABP model was compared with the unoptimized BP diagnosis model,which showed that the modeling time of optimized BABP model was reduced to 0. 1976 s from 5. 1844 s,the positive diagnosis rate of deficiency syndrome was increased to 89. 1809% from 76. 4567 and that of excess syndrome was increased to 70. 1170% from 64. 8441%. Simulation analysis showed that the general ability and classification efficiency of GABP-based neural network are better than those of traditional BP neural network,and can thus provide a better computer network model for the differential diagnosis of large intestine carcinoma syndromes in clinical practice.