采用随机森林方法的声带小结与声带囊肿识别研究  

Research of Vocal Nodules and Vocal Cyst Recognition by Random Forest Method

作　　者：徐天琪[1] 周婉艺 葛怡雯 冯可瞧 陶智[1] 张晓俊[1] 

机构地区：[1]苏州大学物理与光电.能源学部,苏州215006

出　　处：《信息化研究》2017年第1期37-41,共5页INFORMATIZATION RESEARCH

基　　金：国家自然科学基金(No.61271359);江苏省高等学校大学生创新创业训练计划项目(No.201610285015)

摘　　要：声带小结和声带囊肿作为喉部疾病中较为常见且临床症状相似的两种病症,如完全依靠医师的临床经验判断,误诊率较大。文章采用声学检测的方法,提出了一种基于决策树改进的随机森林识别算法对声带小结和声带囊肿进行识别研究。首先对声带小结和声带囊肿各参数进行相关性分析,得到各组参数的Pearson系数用以判断参数的相关性大小,进行声学参数的优化,随后采用随机森林方法进行识别研究。实验数据来源于苏州大学病理嗓音数据库。研究表明,该方法适用于声带小结和声带囊肿的识别,且最终达到81%的识别率。As common laryngeal, vocal nodule and vocal cyst have similar clinical symptoms. If doctors entirely rely on clinical experience, the misdiagnosis rate is high. This paper used a detection algorithm based on random forest method which is a improved decision tree method to recognize vocal nodule and vocal cyst. Firstly, correlation analysis of various parameters about these two diseases was conducted and the associated Person coefficients were acquired to judge the correlation size of parameters for optimizing acoustic parameters. Then with random forest method the research on identification of these parameters was completed. This identi- fication experiment was based on the pathological voice database from Soochow university. Through the research , random forest method is applicable to vocal nodule and vocal cyst recognition and finally achieved recog-nition rate of 81%.

关 键 词：声带小结 声带囊肿 参数优化 随机森林 

分 类 号：TN912.34[电子电信—通信与信息系统]