基于局部-全局自适应信息学习的脑肿瘤磁共振图像分割  

Image Segmentation of Brain Tumor Magnetic Resonance Based on Local-global Adaptive Information Learning

作　　者：陈进杨 王雪真 洪金省[2] 钟婧 时鹏[1,4] CHEN Jin-Yang;WANG Xue-Zhen;HONG Jin-Sheng;ZHONG Jing;SHI Peng(College of Computer and Cyber Security,Fujian Normal University,Fuzhou 350117,China;Radiotherapy Department of the First Affiliated Hospital,Fujian Medical University,Fuzhou 350001,China;Department of Radiology,Fujian Cancer Hospital,Fujian Medical University,Fuzhou 350014,China;Digital Fujian Environmental Monitoring Internet of Things Laboratory,Fujian Normal University,Fuzhou 350117,China)

机构地区：[1]福建师范大学计算机与网络空间安全学院,福州350117 [2]福建医科大学附属第一医院放疗科,福州350001 [3]福建医科大学福建省肿瘤医院放射科,福州350014 [4]福建师范大学数字福建环境监测物联网实验室,福州350117

出　　处：《计算机系统应用》2022年第4期59-67,共9页Computer Systems & Applications

基　　金：福建省科技创新联合资金项目(2018Y9112);福建省卫生健康科研人才培养项目(2019-ZQN-17)。

摘　　要：针对脑肿瘤磁共振成像(MRI)勾画数据少、类别不平衡以及各个私有的数据库具有较大差异导致脑肿瘤MRI图像分割困难的问题,提出了一种基于局部-全局自适应信息学习(ALGIL)分割算法.该方法只需要少量的勾画数据,解决了传统监督学习中对勾画数据数量的依赖问题.通过融合图像的空间域信息和频域信息,利用小波变换将图像从空间域转换到频域,从低频和高频子带中分别提取统计特征和纹理特征,解决了传统单一领域特征提取的局限性;利用局部-全局自适应信息学习算法,首先通过随机森林算法得到特征权重对图像进行赋权并构造相似性矩阵,然后利用指数衰减函数自适应调整标注样本对算法的影响程度,解决了因勾画数据少导致分割不理想的问题.所提方法在公开数据集Brats2018上的实验结果显示,该方法与其他先进模型相比,各项评价指标均有所提升,并且减少了对勾画样本量的需求,大幅提高了图像分割的效率,为脑胶质瘤的自动精确分割提供了新的思路.Segmentation of brain tumors in magnetic resonance imaging(MRI) is a difficult job due to the delineation data shortage, class imbalance, and significant differences among private databases. To solve those problems, we propose an adaptive local-global information learning(ALGIL) segmentation algorithm. This method only requires a small amount of delineation data, which solves the problem of traditional supervised learning depending on the amount of delineation data.The spatial-domain and frequency-domain information of the image is fused to convert the image from the spatial domain to the frequency domain through wavelet transform, and statistical features and texture features are extracted from the low-frequency and high-frequency sub-bands respectively. The limitations of traditional single-domain feature extraction are thereby resolved. With the ALGIL segmentation algorithm, we construct a similarity matrix by weighting the image according to the feature weights obtained through the random forest algorithm. Then, an exponential decay function is employed to adaptively adjust the degree of influence of the labeled samples on the algorithm. In this way, we solve the problem of poor segmentation results caused by the delineation data shortage. The experimental results of the proposed method on the public data set Brats2018 show that compared with other advanced models, the method has greatly improved the efficiency of image segmentation by enhancing various evaluation indicators and reducing the need for delineation samples. It provides a new idea for the automatic and accurate segmentation of brain gliomas.

关 键 词：脑胶质瘤 监督学习 小波变换 纹理特征 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] R739.41[自动化与计算机技术—计算机科学与技术]