癌前细胞诊断的特征选择 Feature Selection for Pre-Cancer Cell Diagnosis 梅建新 1 2 first-author 王思贤 1 2 武汉大学测绘遥感信息工程国家重点实验室 武汉大学测绘遥感信息工程国家重点实验室 武汉大学电子信息学院,湖北,武汉,430072 武汉大学电子信息学院,湖北,武汉,430072 指出了目前基于 DNA- Index的定量细胞学诊断系统中对于癌前非典型增生细胞诊断的不足 ,提出了运用模式识别的方法对癌前非典型增生细胞是否存在恶变倾向进行有效识别 .针对分类识别过程中 ,细胞特征参量过多 ,分类器设计困难的具体情况 ,采用了基于 Tabu Search的搜索策略来解决识别特征的选取问题 .实验表明 :通过选择适当的搜索参数 ,基于 Tabu Search的搜索方法无论是在搜索效率还是在结果的优化程度上都明显地优于其它传统的方法 .搜索得到的特征参量能够有效地描述癌前增生细胞的不同性状 ,这在癌前的早期诊断方面具有良好的实用性 . At the present time,some advanced cytology quantitative detection system can recognise cancer cells accurately through some feature parameter such as DNA-Index.But less attention is put upon prophase of cell cancerization-dysplasia.In this paper,a new method using pattern recognition is put forward to solve this problem.Based on the situation that there are too many feature parameters for classification,a new search strategy,Tabu Search,is put forward to solve the problem of feature selection.Result shows that through selecting eggective searching parameters,the Tabu Search is superior to other traditional selection methods and a group of features can be selected to classified different hyperplasic cells.The method has a good prospect in pre-cancer cell diagmosis. 肿瘤诊断 特征选择 模式识别 Tabu Search tumor diagnosis feature selection pattern recognition Tabu Search R730.4 国家“九五”科技攻关项目 ( 96 -90 1-0 7-0 3);武汉市科委 2 0 0 0年科技重大攻关项目 ( 2 0 0 0 6 0 0 10 10 ) 2001-06-01 2021-04-01 6