基于磁共振图像特征的脑胶质瘤MIB-1指数无创检测  

Noninvasive Prediction of Glioma MIB-1 Index Based on MRI Feature

作　　者：潘颖[1] 刘晨彬[1] 张海石[2] 黄峰平[2] 夏顺仁[1] 

机构地区：[1]浙江大学生物医学工程教育部重点实验室,浙江杭州310027 [2]复旦大学附属华山医院,上海200040

出　　处：《航天医学与医学工程》2012年第3期207-211,共5页Space Medicine & Medical Engineering

基　　金：国家自然科学基金项目(81101903;60772092);浙江省自然科学基金项目(Y1100219)

摘　　要：目的探讨脑胶质瘤磁共振图像特征与MIB-1指数的关系。方法利用基本灰度信息、灰度共生矩阵、灰度-梯度共生矩阵、游程长度矩阵和闵可夫斯基泛函来构建磁共振图像肿瘤区域的原始特征集,进而分别利用基于顺序后退与k-最邻近的方法(SBS-KNN)和基于离散粒子群与支持向量机的方法(DPSO-SVM)对原始特征集进行优化,最后利用优化后的特征集进行分类。结果采用DPSO-SVM方法优化的特征集能有效地预测MIB-1指数的范围,在T1加权序列上准确率达到80.88%。结论磁共振图像特征与MIB-1指数密切相关。本文所提出的算法可以较好地预测出MIB-1指数的范围。Objective To study the correlation between MRI feature and MIB-1 index of glioma. Methods By extracting MRI features, such as basic gray information, gray-level co-occurrence matrix（GLCM） , gray levelgradient co-occurrence matrix（ GGCM）, run-length matrix （RLM） and Minkowski functional, the initial feature set was constructed. Then, the initial feature set was optimized by sequential backward selection combined with k-nearest neighbor （SBS-KNN） and discrete particle swarm optimization combined with support vector machine （DPSO-SVM） , respectively. In the end, the optimized feature set was used in categorization. Results The feature set optimized by DPSO-SVM on Tl-weighted images can efficiently predict the range of MIB-1 and achieve the identification rate of 80.88%. Conclusion MIB-1 index correlates strongly with features of MR images. The proposed algorithm can predict the range of MIB-1.

关 键 词：脑胶质瘤 MIB-1指数 特征选择 支持向量机 无创检测 

分 类 号：R445.2[医药卫生—影像医学与核医学] TP391[医药卫生—诊断学]