基于集成改进ELM的甲状腺疾病分类方法  

Thyroid Disease Classification Method Based on Integration Improved ELM

作　　者：赵杰[1] 张振宇[1] 门国尊[2] 

机构地区：[1]河北大学电子信息工程学院,河北保定071000 [2]河北大学经济学院,河北保定071000

出　　处：《计算机仿真》2014年第3期392-396,共5页Computer Simulation

基　　金：河北省卫生厅科研基金资助项目(20120395)

摘　　要：研究针对B超图像的甲状腺疾病分类问题。甲状腺疾病的计算机分类是提高甲状腺疾病诊断效率的重要途径,包括特征提取和分类器实现,传统方法中特征提取不全面和使用单个分类器,使得诊断精度偏低且结果稳定性差。针对上述问题,提出一种基于集成改进极端学习机的甲状腺疾病分类方法。首先,分析甲状腺B超图像,对临床鉴别甲状腺结节良恶性的特征进行量化,提取了紧致度等9个特征作为数据集;而后,将聚类思想与极端学习机方法融合,结合k-means聚类算法,提出一种新的聚类标准,对数据集进行聚类;最后,对聚类后的子集进行分类训练,并采用多数投票的策略对子分类器进行集成。实验结果表明,改进方法在分类精度和稳定性上较传统算法均有较大提高。Research for thyroid disease classification method based on B-ultrasonic images. Computer classifica- tion of thyroid disease is an important way to improve the efficiency of thyroid disease diagnosis, including feature ex- traction and classification. An extreme learning machine ensemble method based on improved clustering analysis was proposed in the paper. First, through the analysis of the B-ultrasonic images, we quantified the characteristics of the clinical differentiation between benign and malignant thyroid nodules, extracted 9 features such as compactness as the data set. This algorithm integrates the clustering ideology into extreme learning machine, combined with the k- means clustering algorithm, and proposes a new clustering standard. In this method, the samples were clustered into k subsets. At last, extreme leaning machine was used to train these k different subsets, and the ELMs were grouped into an ensemble classifier by the strategy of majority voting. The experimental results show that this method has bet- ter classification accuracy and stability.

关 键 词：极端学习机 聚类分析 集成 甲状腺疾病 

分 类 号：TN391[电子电信—物理电子学]