摘要
针对乳腺肿瘤的诊断率及精准度较低的情况,提出一种基于改进的矢量量化(LVQ)神经网络乳腺肿瘤诊断算法。首先,基于LVQ1算法和LVQ2算法在网络训练过程中更新神经元数目的不同,建立结合LVQ1算法和LVQ2算法的复合LVQ神经网络;然后,考虑到不同的竞争层节点数对LVQ神经网络诊断率的影响,采用K交叉验证法确定复合LVQ最佳网络结构;最后,探讨了不变的学习率在网络训练后期对收敛速度的影响,采用自适应速率算法调整学习率,减少迭代次数。以Wisconsin Breast Cancer Database为实验样本,运用改进算法构造乳腺肿瘤与症状之间的非线性映射关系,用混淆矩阵的概念表达算法诊断准确率。实验结果表明,提出的改进算法诊断准确率达97.1%,相比LVQ1算法和LVQ2算法,误诊率分别降低了5.8%和2.9%。
In view of the low diagnosis rate and accuracy of breast tumors,a breast tumor diagnosis algorithm based on the improved learning vector quantization(LVQ)neural network is proposed. Since the LVQ1 algorithm and LVQ2 algorithm have different neuron updating number in the process of network training,a compound LVQ neural network combining the LVQ1 algorithm and LVQ2 algorithm is established. And then,considering the influence of different number of nodes in competing layers on the diagnostic rate of LVQ neural network,the K cross validation method is used to determine the optimal network structure of the compound LVQ. The influence of constant learning rate on the convergence rate in the later stage of network training is discussed. The adaptive rate algorithm is adopted to adjust the learning rate and reduce the iterations. The Wisconsin Breast Cancer Database is taken as the experimental sample. The improved algorithm is used to construct the non-linear mapping relationship between Breast Cancer and symptoms. The concept of confounding matrix is used to express the algorithm ′ s diagnostic accuracy rate. The experimental results show that the diagnostic accuracy of the proposed improved algorithm is97.1%,and its misdiagnosis rate is reduced by 5.8% and 2.9% respectively in comparison with LVQ1 algorithm and LVQ2 algorithm.
作者
赵小强
张莺莺
ZHAO Xiaoqiang;ZHANG Yingying(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Key Laboratory of Gansu Advanced Control for Industrial Processes,Lanzhou University of Technology,Lanzhou 730050,China;National Experimental Teaching Center of Electrical and Control Engineering,Lanzhou University of Technology,Lanzhou 730050,China)
出处
《现代电子技术》
2022年第1期77-82,共6页
Modern Electronics Technique
基金
国家自然科学基金项目(61763029)
国家自然科学基金项目(61873116)
国防基础科研项目(JCKY2018427C002)。
关键词
乳腺肿瘤诊断
改进LVQ神经网络
K交叉验证法
自适应速率
混淆矩阵
算法改进
breast tumor diagnosis
improved LVQ neural network
K cross validation method
adaptive rate
confusion matrix
algorithm improvement
