基于集成随机森林模型的肺结节良恶性分类  

Classification of benign and malignant pulmonary nodules based on ensemble random forests

作　　者：胡会会 龚敬[1] 聂生东[1] Hu Huihui;Gong Jing;Nie Shengdong(School of Medical Instrument&Food Engineering,University of Shanghai for Science&Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学医疗器械与食品学院,上海200093

出　　处：《计算机应用研究》2018年第10期3117-3120,3125,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(60972122);上海市自然科学基金资助项目(14ZR1427900)

摘　　要：针对目前计算机辅助肺结节良恶性分类模型精度较低的问题,提出了一种基于CT图像的集成随机森林模型肺结节良恶性鉴别方法。首先分割肺结节区域,提取其影像学特征向量输入多个基分类器;然后利用每个基分类器的置信度构建集成模型的分类损失函数,求出每个基分类器的权重;最后根据每个基分类器输出的类别概率值进行加权求和,求得其中概率最大值的类作为分类类别。为验证提出的分类模型性能,设计三种实验方案进行测试,准确率分别达到96.41%、91.36%、95.82%;与已有的肺结节良恶性分类模型进行对比,结果表明,集成随机森林分类模型能够有效提高肺结节鉴别良恶性的准确度。To solve the problem of relatively low accuracy of present classification models,this paper proposed an integrated random forest classification model for malignant-benign pulmonary nodules.Firstly,this method segmented the lung nodule and extracted nodule features as the input of multiple base classifiers.Then,it constructed the classification loss function of the integrated model using the confidence of each base classifier,and it could obtain the weights of each base classifier.Finally,for the unclassified samples,it obtained the weighted sum of the class probability values of each class,and selected the class of the maximum probability as the decision value.To verify the performance of the proposed classification model,this paper set three experiment schemes and the corresponding accuracies were 96.41%,91.36%and 95.82%,respectively.At the same time,comparing with other classification methods,the results show that the proposed classification model can effectively improve the accuracy of malignant-benign pulmonary nodules classification.

关 键 词：计算机辅助诊断 CT图像 肺结节良恶性分类 集成随机森林 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]