一种基于树搜索的层次多标签乳腺疾病分类诊断方法  

Classification and prediction method of hierarchical multi-label breast disease based on tree search

作　　者：金程笑 潘乔[1] 张敬谊[2] 俞春儒 JIN Chengxiao;PAN Qiao;ZHANG Jingyi;YU Chunru(School of Computer Science and Technology,Donghua University,Shanghai 201620,China;Wonders Information Co.,Ltd.,Shanghai 201112,China)

机构地区：[1]东华大学计算机科学与技术学院,上海201620 [2]万达信息股份有限公司,上海201112

出　　处：《智能计算机与应用》2020年第2期34-39,共6页Intelligent Computer and Applications

基　　金：上海市经信委人工智能创新发展专项资金(RX-RJJC-08-16-0483,2017-RGZN-01004)。

摘　　要：随着医疗信息化的快速发展,医疗机构在临床诊断的过程中产生了大量的原始电子病历数据,存在着大量的可挖掘信息,作为临床的辅助诊断。由于乳腺疾病患者的患病情况较为复杂,同一位患者可能会患有多种相关疾病,每个大类疾病分类下可能会存在很多的小类疾病,而小类疾病分类下又可能存在更细粒度的疾病类别。传统的分类问题(如二分类和多标签分类)往往会忽略各标签之间存在的依赖关系并且分类算法输出数目呈指数级,占用空间过大,造成预测性能不佳。因此本文提出了一种基于树搜索的层次多标签乳腺疾病分类诊断方法,利用树结构可以充分考虑到标签集之间的层次结构的依赖关系,规范化诊断结论。按诊断结果之间的层次关系构建了层次多标签树,通过对标签树的路径搜索,最终实现乳腺疾病的多标签分类。With the rapid development of medical informatization,medical institutions generate a large amount of original electronic medical record data during the clinical diagnosis process,and there is a large amount of information that can be mined for clinical auxiliary diagnosis.Because the prevalence of patients with breast disease is more complicated,the same patient may suffer from multiple related diseases.There may be many small diseases under each major disease classification,and there may be more detailed diseases categories under the small disease classification.In the traditional classification problems(such as binary classification and multi-label classification),the dependencies between the labels are tended to be ignored and the number of classification algorithm outputs is exponential,taking up too much space,resulting in poor prediction performance.Therefore,a hierarchical multi-label breast disease classification diagnosis method based on tree search is proposed in this paper.The tree structure can fully consider the hierarchical relationship between label sets and standardize the diagnosis conclusion.According to the hierarchical relationship between the diagnosis results,a hierarchical multi-label tree is constructed.By searching the path of the label tree,the multi-label classification of breast diseases is finally realized.

关 键 词：乳腺疾病 树搜索 层次多标签 规范化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]