摘要
为了对肺结节的良、恶性诊断形成定量的客观分析和提高良、恶性的分类正确率,针对肺结节CT图像提出了一种基于Bootstrap-异质SVM的集成学习方法.首先,采用模糊聚类图像分割方法提取肺结节,计算提取出的结节特征参数用于学习分类.然后,以支持向量机(SVM)在不同核函数下的不同性能构造高差异性的子学习器,在子学习器中引入Bootstrap算法来提高其学习精度,通过集成学习方法实现学习器分类性能的整体改善.对146个(40个良性,106个恶性)肺结节样本分别利用单个SVM、BP神经网络和Bootstrap-异质SVM集成学习方法进行了学习测试,获得的最高分类正确率分别为80%,、82%,和90%,.实验结果表明:提出的Bootstrap-异质SVM集成学习方法将单个SVM分类器的最高正确率提高了10%,,同时也获得了高于BP神经网络8%,的分类正确率和较好的学习稳定性,有效地改善了机器学习在不平衡数据集下对肺结节良恶性的分类能力.
In order to diagnose lung nodules quantitatively and objectively and increase the classification accuracy,this study presents a classification method for lung nodules on CT images by ensemble learning based on Bootstrapheterogeneous SVM.Firstly,a semi-automated segmentation method based on fuzzy clustering method was used to extract lung nodule pictures from lung CT images. The characteristic parameters were abstracted from lung nodule pictures for learning and classification.Next,the ensemble classifier model based on Bootstrap-heterogeneous SVM was constructed by high precision and high otherness sub-classifiers produced by Bootstrap algorithm and SVMs with different kernel functions to improve the overall performance of classification.In the experiment,single SVMs and BP Neural Network and the ensemble learning method based on Bootstrap-heterogeneous SVM were respectively used on 146(40 benign and 106 malignant)lung nodule samples to diagnose lung nodules.The highest classification accuracy is 80%,,82%,,90% respectively.The experimental results show that the ensemble learning method based on Bootstrap-heterogeneous SVM increases the highest classification accuracy of single SVM classifier by 10%,and it is also superior to BP Neural Network in the classification accuracy with an increase of 8%, as well as in stability of learning.The proposed method effectively improves the classification performance of machine learning about lung nodules under unbalanced data sets.
作者
高峰
代美玲
祁瑾
Gao Feng;Dai Meiling;Qi Jin(School of Precision Instrument and Opto-Electronics Engineering,Tianjin University,Tianjin 300072,China;Tianjin Key Laboratory of Biomedical Detecting Techniques and Instruments,Tianjin 300072,China;Cancer Institute and Hospital,Tianjin Medical University,Tianjin 300060,China)
出处
《天津大学学报（自然科学与工程技术版）》
EI
CSCD
北大核心
2017年第3期321-327,共7页
Journal of Tianjin University：Science and Technology
基金
国家自然科学基金资助项目(61575140)~~
