中文摘要：

乳腺组织疾病早期诊断具有十分重要的意义，基于电阻抗频谱法进行乳腺组织疾病早期诊断是近年来发展起来的乳腺临床检查技术，具有成本低、无损害、无创伤等优势。本研究采用电阻抗频谱法测量的64位妇女，106个乳腺样本的电阻抗特性数据，基于神经网络与支持向量机实现乳腺组织疾病的早期自动诊断。针对传统神经网络的过拟合问题，网络初始参数如何确定问题，以及支持向量机（SVM）分类器的惩罚因子C与核函数g如何有效确定问题，采用遗传算法全局寻优的方式，按适者生存的原则，进行参数的选择、交叉、变异，直到收敛到最适应环境的“染色体”上，即得到问题的最优解，从而实现网络权值与阈值以及SVM惩罚因子与核函数的参数寻优。改进后的算法诊断识别率分别由优化前的61．92％和51．92％，提高到优化后的76．15％和68．08％，为乳腺组织疾病的早期自动诊断提供了一种有效的参考方法。

英文摘要：

It is very important to make early diagnosis with breast tissue disease. The method based on Electrical Impedance Spectroscopy is effective for the clinical diagnosis. Advantages of the method include low- cost and non-damage. In this paper, we focused on the early automatic diagnosis of breast tissue disease based on two intelligent classifiers ; BP network and support vector machine （ SVM）. The data of Electrical Impedance Spectroscopy come from 64 subjects, including 106 breast tissue samples. Focus on the uncertain property of selecting characters of the classifiers, parameter optimization and selecting were conducted based on the Genetic Algorithm. Based on principle of the overall optimization and the survival of the fittest, selection-crossover- mutation was done to pursue the optical parameters. After the process, the accuracy of automatic diagnosis was improved from 61.9% and 51.9% to 76.2% and 68.0%. The method is expected to provide an effectual method for the clinical automatic diagnosis of breast tissue disease.