Print

发布时间: 2020-03-16
摘要点击次数:
全文下载次数:
DOI: 10.11834/jig.190524
2020 | Volume 25 | Number 3




    学者观点    




  <<上一篇 




  下一篇>> 





MRI脑肿瘤图像分割研究进展及挑战
expand article info 李锵, 白柯鑫, 赵柳, 关欣
天津大学微电子学院, 天津 300072

摘要

脑肿瘤分割是医学图像处理中的一项重要内容,其目的是辅助医生做出准确的诊断和治疗,在临床脑部医学领域具有重要的实用价值。核磁共振成像(MRI)是临床医生研究脑部组织结构的主要影像学工具,为了使更多研究者对MRI脑肿瘤图像分割理论及其发展进行探索,本文对该领域研究现状进行综述。首先总结了用于MRI脑肿瘤图像分割的方法,并对现有方法进行了分类,即分为监督分割和非监督分割;然后重点综述了基于深度学习的脑肿瘤分割方法,在研究其关键技术基础上归纳了优化策略;最后介绍了脑肿瘤分割(BraTS)挑战,并结合挑战中所用方法展望了脑肿瘤分割领域未来的发展趋势。MRI脑肿瘤图像分割领域的研究已经取得了一些显著进展,尤其是深度学习的发展为该领域的研究提供了新的思路。但由于脑肿瘤在大小、形状和位置方面的高度变化,以及脑肿瘤图像数据有限且类别不平衡等问题,使得脑肿瘤图像分割仍是一个极具挑战的课题。由于分割过程缺乏可解释性和透明性,如何将全自动分割方法应用于临床试验,还需要进行深入研究。

关键词

脑肿瘤图像分割; 核磁共振成像(MRI); 监督分割; 非监督分割; 深度学习

Progresss and challenges of MRI brain tumor image segmentation
expand article info Li Qiang, Bai Kexin, Zhao Liu, Guan Xin
School of Microelectronics, Tianjin University, Tianjin 300072, China
Supported by: Tianjin Municipal Natural Science Foundation(16JCZDJC31100)

Abstract

Brain tumor segmentation is an important part of medical image processing. It assists doctors in making accurate diagnoses and treatment plans. It clearly carries important practical value in clinical brain medicine. With the development of medical imaging, imaging technology plays an important role in the evaluation of the treatment of brain tumor patients and can provide doctors with a clear internal structure of the human body. Magnetic resonance imaging (MRI) is the main imaging tool for clinicians to study the structure of brain tissue. Brain tumor MRI modalities include T1-weighted, contrast-enhanced T1-weighted, T2-weighted, and liquid attenuation inversion recovery pulses. Different imaging modalities can provide complementary information to analyze brain tumors. These four types of modalities are usually combined to diagnose the location and size of brain tumors. At present, due to the extensive application of MRI equipment in brain examination, a large number of brain MRI images are generated in the clinical setting. This rise in the quantity of brain MRI images hinders doctors in manually annotating and segmenting all images promptly. Moreover, the manual segmentation of brain tumor tissues highly depends on doctors' professional experience. Therefore, research has focused on the ways to segment brain tumors efficiently, accurately, and automatically. In recent years, significant advancement has been made in the study of brain tumor segmentation methods. To enable other researchers to explore the theory and development of segmentation methods for brain tumor MRI images, this work reviews the current research status in this field. In this study, the current semiautomatic and fully automatic segmentation methods for brain tumor MRI images are divided into two categories:unsupervised segmentation and supervised segmentation. The difference between the two methods lies in the use of hand-labeled image data. Unsupervised segmentation is a nonpriori image segmentation method based on computer clustering statistical analysis. The unsupervised methods are divided into threshold-based, region-based, pixel-based, and model-based segmentation technologies according to the different segmentation principles. This work briefly describes the unsupervised methods according to the above classification and summarizes their advantages and disadvantages. The main feature of supervised segmentation is its use of labeled image data. The segmentation process involves model training and testing. In the former, labeled data are used to learn the mapping from image features to labels. In the latter, the model assigns labels to unlabeled data. Supervised segmentationis mainly based on the segmentation technology of pixel classification. It generally includes traditional machine learning methods and methods based on neural networks. The common traditional machine learning methods and the methods based on neural networks used in brain tumor segmentation are briefly described herein, and their advantages and disadvantages are summarized. The segmentation methods for brain tumor images based on deep learning are mainly described. With the advancement of artificial intelligence, deep learning, especially the new technology represented by convolutional neural networks (CNNs), has been well received because of its superior brain tumor segmentation results. Compared with traditional segmentation methods, CNNs can automatically learn representative complex features directly from the data. Hence, the research of brain tumor segmentation based on CNNs mainly focuses on network structure design rather than image processing before feature extraction. This study focuses on the structure of neural networks used in the field of brain tumor image segmentation and summarizes the optimization strategies of deep learning. Lastly, the challenge of brain tumor segmentation (BraTS) is introduced, and the future development trend of brain tumor segmentation is established in combination with the methods used in the challenge. The BraTS challenge is a competition to evaluate the segmentation methods for brain tumor MRI images.The BraTS challenge uses preoperative MRI image data from multiple institutional brains to focus on the segmentation of gliomas. In addition, the BraTS challenge involves predicting overall patient survival by combining radiological features with machine learning algorithms to determine the clinical relevance of this segmentation task. The segmentation methods for brain tumor MRI images have inherent advantages, disadvantages, and application scope. Researchers have been working on how to improve the accuracy of segmentation results, the robustness of models, and the overall operational efficiency. Hence, this study analyzes the advantages and disadvantages of various methods, the optimization strategies of deep learning, and future development trends. The optimization strategy of deep learning is as follows.In the aspect of imaging data, data enhancement techniques, such as flipping, scaling, and cropping, are used to increase the amount of training data and improve the generalization ability of models. Acascade framework is introduced to realize the segmentation of whole tumors, core tumors, and enhanced tumors by combining the acascade framework with the inclusion relationship of tumors in the brain anatomical structure. An improved loss function is used to deal with image category imbalances. In terms of network structure, multiscale and multichannel strategies are adopted to make full use of image feature information.In the process of downsampling, aconvolution operation is used instead of pooling so that image information can be further learned while reducing image information loss.Between the convolutional layers, the jump connection method is applied to effectively solve the degradation problem of the deep network.In different cases, the appropriate standardization method, activation function, and loss rate are selected to achieve satisfactory segmentation effects. This work summarizes the development trend of brain tumor segmentation methods by learning and arranging the methods used in the BraTS challenge. As a result of the diversification of MRI imaging modalities, making full use of the each modal image information can effectively improve the accuracy of brain tumor segmentation. Therefore, the reasonable utilization of multimodality images can be expected to become a research hotspot.Methods based on deep learning are outstanding in the field of brain tumor segmentation and have become a hot research direction.The defects of machine learning algorithms lead to an inaccurate segmentation of brain tumors. A popular trend is to improve the original method or combine various methods effectively. Remarkable progress has been made in the segmentation of brain tumor MRI images. The development of deep learning, in particular, provides new ideas for the research in this field. However, brain tumor image segmentation is still a challenging subject because brain tumors vary in size, shape, and position. Moreover, brain tumor image data are limited, and the categories are not balanced. As a result of the lack of interpretability and transparency in the segmentation process, the application of a fully automated segmentation method to clinical trials still requires further research.

Key words

brain tumor image segmentation; magnetic resonance imaging (MRI); unsupervised segmentation; supervised segmentation; deep learning

0 引言

脑肿瘤是指在大脑内部生长的不正常细胞群(DeAngelis,2001)。脑肿瘤可分为良性肿瘤和恶性肿瘤,良性脑肿瘤可通过手术治愈,恶性脑肿瘤又称为脑癌,是最致命的癌症类型之一,会直接导致患者死亡(Siegel等,2013)。全球癌症发病率和死亡率正在迅速增长,根据《临床医师癌症杂志》 CA(A Cancer Journal for Clinicians)全球癌症统计报告(Bray等,2018),2018年脑癌新发病例约29.7万,约占所有癌症新发病例的1.6 %,脑癌死亡病例约24.1万,约占所有癌症死亡病例的2.5 %。

成人中最常见的脑肿瘤是原发性中枢神经系统淋巴瘤和胶质瘤,其中胶质瘤占恶性肿瘤的80 %以上(Schwartzbaum等,2006)。根据胶质瘤的侵袭性,大致将其分为高级别胶质瘤HGG(high-grade Gliomas)和低级别胶质瘤LGG(low-grade Gliomas)。HGG是恶性肿瘤,生长迅速且易形成异常组织,HGG患者死亡率高,五年存活率仅为5.5 % (Bauer等,2013);LGG可以是良性或恶性,生长速度较慢,治疗后可能复发并发展成HGG。因此,胶质瘤是脑肿瘤分割的重点对象。

随着医学成像的发展,成像技术在脑肿瘤患者治疗评估中发挥重要作用,能为医生提供清晰的人体内部结构。脑部医学领域中常见的成像技术有:计算机断层成像、单光子发射计算机断层成像、正电子发射断层扫描成像、磁共振波谱成像和核磁共振成像MRI(magnetic resonance imaging)等,这些成像技术能够提供脑肿瘤形状、大小和位置信息,为医生做出准确的诊断和制定治疗方案提供条件。但由于MRI是一种非侵入性成像模式,具有良好的软组织对比度,无放射线损害,无骨性伪影,能多方面多参数成像等独特的优点(Liang和Lauterbur,2000)。因此,MRI特别适合用于临床脑部病变检查。

脑肿瘤MRI模态包括T1加权(T1-weighted)、对比增强T1加权T1C(contrast enhanced T1-weighted)、T2加权(T2-weighted)和液体衰减反转恢复脉冲FLAIR(fluid attenuated inversion recovery)等成像模态,不同成像模态可以提供补充信息来分析脑肿瘤信息,临床上通常结合以上4种图像共同诊断脑肿瘤的位置和大小(Prastawa等,2004)。图 1显示了同一患者脑部的4种MRI模态及其对应的专家分割结果图(Havaei等,2017)。图 1(a)是T1成像模态,因操作简单常用于脑肿瘤的结构分析,但成像质量差,无法提供更为详尽的肿瘤特征信息;图 1(b)是T2成像模态,水肿区域相比其他成像方式更加明亮,缺点是脑脊液和肿瘤的像素特征难以区分;图 1(c)是T1C成像模态,由于增生性脑肿瘤区血脑屏障破坏导致的造影剂积聚,使得T1C成像中的脑肿瘤边界变得更明亮,很容易区分出肿瘤和囊变区域;图 1(d)是FLAIR成像模态,水肿区域边界明显,是目前分割脑肿瘤比较有效的成像方式;图 1(e)是专家分割结果图,绿色区域为由坏死、水肿、非增强性肿瘤和增强肿瘤构成的全肿瘤,黄色区域为由坏死、非增强肿瘤和增强肿瘤构成的肿瘤核心,红色区域为增强肿瘤(Corso等,2008)。目前对脑肿瘤的分割不限于全肿瘤,还包括肿瘤核心和增强肿瘤等子结构的分割。

图 1 脑肿瘤患者MRI的4种模态及专家分割结果图
Fig. 1 Four models of brain tumor MRI images and the expert segmentation result
((a) T1; (b) T2; (c) T1C; (d) FLAIR; (e) expert segmentation result map)

由于MRI设备在脑部检查中的广泛应用,临床中会产生大量脑部MRI图像数据,医生不可能及时手动注释和分割全部图像,并且人工手动分割脑肿瘤组织依赖于医生的个人经验。因此,如何高效、精准且全自动地分割脑肿瘤成为研究重点。对脑肿瘤分割方法的研究取得了重大的进展。通常情况下,脑肿瘤分割方法容易检测到异常组织,但仍缺少准确、可重复且适应性强的脑肿瘤分割算法。

本文跟踪基于MRI图像的脑肿瘤分割领域有代表性和前沿的论文,综述了各种方法的研究进展、优缺点和结果评价;重点综述了目前最流行的基于深度学习的脑肿瘤分割方法,并将其优化策略进行了归纳;最后介绍了脑肿瘤分割比赛,结合比赛情况总结了该领域的发展趋势。

1 脑肿瘤分割方法概述

根据人与计算机交互作用和技术的发展(Olabarriaga等,2001Gordillo等,2013),脑肿瘤分割方法通常分为3大类:手动分割、半自动分割和全自动分割。

脑肿瘤手动分割涉及手动绘制肿瘤边界和标记感兴趣结构,专家需要掌握脑肿瘤图像信息以及诸如解剖学之类的专业知识(Pham等,2000)。目前,手动分割仍然广泛应用于临床试验。在实践中,专家必须逐片浏览MRI扫描仪生成的多幅2维横断面图像,从中仔细描绘脑肿瘤区域。这是一项耗时费力的任务,而且手动分割结果会因专家经验的不同出现明显差异(Luo等,2003),因此需要半自动和全自动分割方法来解决这些问题。

脑肿瘤半自动分割系统主要由用户界面、人机交互以及软件计算组成(Olabarriaga等,2001),通常需要人工干预进行初始化和检查结果的准确性,甚至需要人为手动校正分割结果。尽管脑肿瘤半自动分割可以获得比手动分割更好的结果,但人工干预同样会对分割结果造成影响。目前大多数针对半自动分割方法的研究目的都是尽可能减少人机的交互作用。因此,提出了全自动脑肿瘤分割方法。

脑肿瘤全自动分割结合了人工智能和先验知识,无需任何人为干预。但由于分割过程缺乏可解释性和透明性(Papageorgiou等,2008),全自动分割目前主要限于研究环境,在临床实践中尚未被医学从业者广泛接受。随着机器学习算法的发展,人工智能可以模拟人类学习能力,脑肿瘤全自动分割已成为一个热门的研究课题。

由于科学研究和临床应用的需要,研究人员提出了多种MRI图像脑肿瘤分割方法,但因MRI固有噪声、伪影等影响以及脑肿瘤在颅内的复杂性,使脑肿瘤半自动和全自动分割面临巨大挑战。本文将目前基于MRI脑肿瘤半自动和全自动分割方法分为两大类:非监督分割和监督分割,两者区别在于是否使用已手动标记的图像数据。

1.1 非监督分割

非监督分割是通过计算机对图像进行集聚统计分析的无先验图像分割方法。使用非监督方法对脑肿瘤进行分割的主要缺点有:分割区域的数量需要预先确定,脑肿瘤图像没有明确定义的灰度强度和边界。因此,在使用非监督方法分割脑肿瘤之前,通常需要对MRI图像进行强度不均匀校正(Madabhushi和Udupa,2005Hou,2006Vovk等,2007)和颅骨剥离(Capelle等,2004)等预处理操作。

非监督方法按照分割原理的不同主要分为:基于阈值的分割技术、基于区域的分割技术、基于像素分类的分割技术以及基于模型的分割技术(Gordillo等,2013Wong,2005Farag等,2005Pham等,2000),根据以上分类对非监督方法进行概述,并将所述方法的优缺点总结于表 1

表 1 常用脑肿瘤分割方法优缺点总结
Table 1 Summary of advantages and disadvantages of common brain tumor segmentation methods

下载CSV
分割方法 监督与否 优点 缺点
基于阈值 全局和局部阈值 简单,计算速度快。 对增强肿瘤区域的适用性有限(Gibbs等,1996)。
基于区域 区域生长 简单,能够正确分割具有相似属性的区域并生成连通区域(Deng等,2010; Węgliński和Fabijań,2011)。 部分容积效应(Sato等,2000),噪声或强度变化可能导致孔洞或过度分割。
分水岭 可以同时划分多个区域,可以生成完整的图像轮廓,并且不需要任何类型的轮廓连接(Dam等,2004)。 容易过度分割(Gies和Bernard,2004)。
基于像素 FCM 始终收敛于脑肿瘤边界。 计算时间长,对噪声敏感(Kannan,2008)。
MRF 能够表示数据实例之间的复杂依赖关系。 控制空间相互作用强度的参数选择较困难,算法复杂度高(Capelle等,2004)。
K-means 简单,运行速度快,实时性好。 K的取值对结果影响太大。
K最近邻 简单,可以有效降噪,易于实施。 对不相关或冗余特性敏感。
基于模型 参数可变形模型 能够适应生物结构随时间和不同个体的差异(McInerney和Terzopoulos,1996)。 在不均匀的情况下,模型可能会收敛到错误的边界(Luo等,2003)。
几何可变形模型 能够捕获多个对象和复杂边界。 速度慢,准确性差。
传统机器学习 SVM 不容易过度拟合,计算复杂度低。 训练较慢,当训练数据不可线性分离时难以确定最优参数并且难以理解算法的结构。
CRF 可以容纳任意的上下文信息,特征设计灵活。 训练代价大、复杂度高。
RF 不容易过度拟合,可以平衡脑肿瘤数据之间的差异,抗干扰能力强。 对于小数据或者低维数据(特征较少的数据),可能不能很好地分割。
神经网络 人工神经网络 能够模拟异常分布和非线性依赖。 脑肿瘤数据集较小,学习阶段较慢(Iftekharuddin等,2009)。
脉冲耦合神经网络 所产生的时间信号对图像的旋转、扩张或平移具有不变性。 分割结果对参数依赖性强。
自组织映射 训练过程较容易,速度较快,常用于高维数据。 可调参数过多,容易产生偏差。
卷积神经网络 特征提取能力强,可直接用于处理原始数据。 分割速度较慢,实时性差。

基于阈值的分割方法通过将图像灰度值与一个或多个强度阈值进行比较,从而实现对图像的分割。根据该强度阈值的不同作用范围,把阈值分割方法分为全局阈值法(Gibbs等,1996)和局部阈值法(Sung等,2000Stadlbauer等,2004)。如果图像灰度直方图为双峰模式,则可以通过称为全局阈值的单个阈值将目标对象与图像中的背景分离;如果图像包含两种以上类型的对象区域,则必须使用多个局部阈值来执行分割。由于脑肿瘤结构的复杂性,阈值分割主要用于确定脑肿瘤的大致位置,通常用作脑肿瘤分割过程的第1步。

基于区域的分割方法通过预定义的相似性标准和均匀性属性来检查图像中的像素,将邻域像素合并,形成不相交的区域(Wong,2005)。区域生长(Rexilius等,2007Deng等,2010Węgliński和Fabijańska,2011)和分水岭(Letteboer等,2004Benson等,2015)是基于区域的方法中最常用的脑肿瘤分割方法。区域生长用于从图像中提取相似像素的连通区域,从至少一个属于感兴趣区域的种子开始,检查种子的邻域,并将满足相似性标准的像素添加到该区域。该过程重复进行,直到不再有像素可以添加到该区域。种子可以手动选择,也可以通过种子自动发现程序提供。分水岭可以生成完整的图像轮廓,并且无需任何类型的轮廓连接,已广泛应用于脑肿瘤分割。

基于像素分类的无监督分割方法通常使用无监督的分类器来对特征空间中的像素进行聚类。图像像素在特征空间中可以使用像素属性来表示。像素属性包括图像中每个像素的灰度级、局部纹理和颜色分量。基于像素分类的无监督分割方法主要包括模糊C均值FCM(fuzzy C-means)(童云飞等,2018任璐等,2018)和马尔可夫随机场MRF(Markov random fields)(Nie等,2009Bauer等,2011b)等。FCM聚类为了缓解区域边界特征过渡不明显的问题,将模糊集的概念引入到分割过程中。MRF聚类考虑到空间信息和图像像素之间的依赖性,将空间信息集成到聚类过程中,减少了可能出现的目标重叠问题以及噪声对聚类结果的影响,从而可以提高脑肿瘤分割任务的准确性。

基于模型的分割方法通过从图像数据获得的约束以及关于脑肿瘤位置、大小和形状的先验知识,为分割目标对象构建连通区域。模型包括参数可变形模型(Shen等,2011)和几何可变形模型(Ho等,2002),参数可形变模型能够适应生物结构随时间和不同个体的显著变化(Tek和Kimia,1995),但其难以处理分割轮廓分裂和合并的拓扑变化,而几何可变模型能够解决该问题(Malladi等,1995)。

1.2 监督分割

监督分割的主要特点是利用了已标记的图像数据。分割过程包括使用已标记数据进行图像特征学习得到图像特征到标签映射的模型训练阶段,以及通过模型将标签分配给未标记数据的测试阶段。监督分割的一个主要优点是可以简单地通过改变训练集来完成不同的任务,但训练集的选择至关重要,使用不同的训练集可能导致训练时间和分割结果的巨大差异(Bezdek等,1993)。

监督方法主要是基于像素分类的分割技术,一般来说,包括传统机器学习和基于神经网络的方法。本文将对脑肿瘤分割中常用的传统机器学习方法和基于神经网络的方法进行概述,所述方法的优缺点见表 1

常见的传统机器学习脑肿瘤分割方法主要有:支持向量机SVM(support vector machine)、条件随机场CRF(conditional random field)和随机森林RF(random forest)。SVM是一种基于参数核的方法,可用来处理监督分类问题(Vapnik,2000),由于其具有良好的分类能力,已被广泛应用于脑肿瘤分割领域(Ruan等,2007Cai等,2007Ruan等,2011Bauer等,2011a)。Bauer等人(2011a)采用SVM算法实现了脑肿瘤的自动分割,但算法的空间和时间复杂度较高,分割效率过低。CRF技术能够表示数据实例之间的复杂依赖关系,进而提高脑肿瘤分割任务的准确性(Rao等,2017)。Wu等人(2014)在CRF框架中使用超像素特征来分割脑肿瘤,但在不同的脑肿瘤患者病例中结果差异太大,特别是在LGG图像中表现更是不佳。RF是属于集成学习的一种组合分类算法,已被证明可大幅度提高训练模型的精度(Lefkovits等,2016)。Pinto等人(2015)基于图像特征采用了极端RF方法将目标和背景进行分类,总体上实现了83%的分割准确率。

基于神经网络的模型主要包括浅层模型和深层模型。大数据时代带来了充足甚至过量的数据支持,且机器计算能力的增长,使优化深层模型成为可能,同时模型的训练逐渐变得可控,因此深层模型成为神经网络的主流。深度学习采用了深层模型,由于其强大的特征学习能力,已广泛应用在脑肿瘤分割领域。

2 基于深度学习的脑肿瘤分割方法

随着人工智能的兴起,深度学习尤其以卷积神经网络CNN(convolutional neural network)为代表的新技术通过展示非常优越的脑肿瘤分割结果而受到欢迎。与传统的分割方法相比,CNN可以直接从数据本身自动学习具有代表性的复杂特征。基于这一特性,CNN脑肿瘤分割方法的研究主要集中在网络架构设计而非特征提取前的图像处理(Işın等,2016)。本节将重点分析脑肿瘤图像分割领域中所使用神经网络的架构,并对深度学习的优化策略进行总结。

2.1 深度学习在脑肿瘤图像分割中的应用

CNN通过堆叠多个卷积层、池化层和全连接层以及将图像与核特征参数进行卷积,进而形成具有强鲁棒性和自适应性的特征学习模型。以CNN中具有代表性的AlexNet网络为例,对CNN结构进行说明,AlexNet网络结构图如图 2所示(Xing,2018)。AlexNet网络由5个卷积层Conv、3个池化层Pool、两个全连接层Fully Connected和1个SoftMax层构成,其中网络中的激活函数用线性整流函数ReLU(rectified linear unit)代替了传统的S或T激活函数,图中C为卷积层卷积核的通道,N为全连接层神经元的数量,Forward代表网络前向传播的特征学习过程,Backward代表网络反向传播的参数学习过程。表 2图 2网络的具体参数。

图 2 AlexNet网络结构示意图
Fig. 2 Network structure chart of AlexNet

表 2 AlexNet网络结构参数
Table 2 Parameters of AlexNet network

下载CSV
卷积核 数量 步长 填充 输出大小
Input Data - - - - 227 × 227 × 3
Conv1 11 ×11 ×3 96 4 0 55 ×55 ×96
Pool1 3 ×3 ×96 - 2 0 27 ×27 ×96
Conv2 5 ×5 ×96 256 1 same 27 ×27 ×256
Pool2 3 × 3 × 256 - 2 0 13 ×13 ×256
Conv3 3 × 3 × 256 384 1 same 13 ×13 ×384
Conv4 3 ×3 ×384 384 1 same 13 ×13 ×384
Conv5 3 ×3 ×384 256 1 same 13 ×13 ×256
Pool3 3 ×3 ×256 - 2 0 6 ×6 ×256
FC1 - - - - 4 096 ×1
FC2 - - - - 4 096 ×1
SoftMax - - - - 1 000
注:“-”代表无数据。

相对于人工分割和传统分割方法,CNN应用在脑肿瘤分割的最大优势在于能够提取脑部MRI图像中肿瘤复杂特征。CNN的特征检测层通过对训练数据的学习,避免了显式的特征抽取,隐式地从训练数据中进行学习,能够大幅度提高脑肿瘤图像分割精度。Havaei等人(2017)提出了一种具有两通路CNN的深度学习模型,包括一个卷积通路和一个全连接通路,但由于网络结构较复杂、模型参数较多导致网络训练较困难,分割精度只能达到85%左右。Kamnitsas等人(2017b)为了结合局部和全局信息,采用双路径结构,可同时处理多个尺度的输入图像,并且将完全连接的CRF模型扩展到3D,用于CNN最终的SoftMax层映射的后期处理,该CRF克服了以前模型的局限性,可以处理任意大的邻近区域,同时加快训练速度。Pereira等人(2016)采用了层数更深的CNN结构,且模型中使用多个3×3小尺寸的卷积核来代替7×7和5×5的大卷积核以提高卷积网络的运算速度,加强脑肿瘤特征的提取,使得分割精度可达87%左右。Dvořák和Menze(2015)将脑肿瘤图像的多分类任务划分为3个二分类子任务,且每个子任务都采用CNN来解决,使得肿瘤核心以及增生性肿瘤的分割精度大幅度提高。师冬丽等人(2018)结合模糊推理系统,通过建立模糊学习规则对CNN预测肿瘤点的概率进行再判断,使得分割精度达到90%左右,但如此一来算法却变为半自动化。

CNN中大部分参数来自全连接层,而全连接层只占据前向运算的一部分,参数数量和运算量不成比例,并且全连接层会打破图像原本的维度。为了克服CNN这些缺点,Long等人(2015)将AlexNet的两个全连接层转化为卷积层,提出了一种像素级别的图像语义分割网络,即全卷积神经网络FCNN(fully convolutional neural network)。FCNN能够端对端地对每个像素进行分类,最后输出仍为2维矩阵,这样保留了像素间的空间信息,更加有利于特征的提取,最终实现图像分割任务。

Chen等人(2016)为了充分利用深层残差学习的强大能力,提出一个深度体素级残差网络,称为VoxResNet,该网络将2维深层残差扩展为3维,并融合了具有深度监督的多层次上下文信息,以进一步提高3D脑肿瘤图像的分割性能。Zhao等人(2017)通过将FCNN和CRF集成,分别在横断面、冠状面和矢状面视图方向上使用具有FCNN参数的图像切片训练CRF,共得到3个分割模型,并且使用基于投票的策略将3个模型融合完成脑肿瘤分割。Chen等人(2018)在FCNN的基础上引入多尺度感受野来进行准确的体素分类,该模型建立在密集连接块上,并利用分层体系结构来考虑不同类型的脑肿瘤,训练过程中使用块级的训练模式以缓解脑肿瘤图像类别不平衡问题。

在生物医学研究领域中,通常无法获得足够多的训练图像。虽然卷积神经网络已被众多研究人员应用在脑肿瘤分割领域,但由于可用训练集较小,脑肿瘤分割精度提高有限。针对此问题,Ronneberger等人(2015)提出了一个用于生物医学图像分割的卷积神经网络U-Net。该网络修改和扩展了FCN结构,使其能够在较少的训练图像下工作,并产生更精确的分割。U-Net网络结构图如图 3所示,该网络结构由捕获上下文的分析路径(左侧)和实现精确定位的合成路径(右侧)组成,每个路径都有5个分辨率层。分析路径和合成路径每层均包含两个内核大小为3×3的卷积层,每个卷积层后跟一个ReLU激活函数。分析路径中相邻两层之间采用内核大小为2×2,步长为2的最大池化层进行下采样,在每个下采样步骤中,将特征通道的数量加倍。合成路径中相邻两层之间采用内核大小为2×2的反卷积层进行上采样,将特征通道的数量减半。之后在合成路径中建立与同层分析路径的快捷连接,为合成路径提供基本的高分辨率特征。在最后一层,使用内核大小为1×1的卷积层将输出通道减少到标签数量。

图 3 U-Net网络结构示意图
Fig. 3 Network structure chart of U-Net

Dong等人(2017)在U-Net训练过程中采用零填充来保证分析和合成路径中所有卷积层的输入输出维数相同,并且为了从原始数据生成更多的训练数据来提高网络性能,使用了大量的数据增强技术。Stawiaski(2017)基于原本U-Net网络结构,分析路径中,在每层均输入该层分辨率大小的图像数据,有效避免了模型训练过程中脑肿瘤特征的丢失;合成路径中,采用多尺度深度监督方式提供更精确的分割结果。Sherman(2018)提出了用V-Net分割MRI脑肿瘤图像,该网络将U-Net扩展到3维,并在同层卷积之间加入残差结构,同时使用步长大于1的卷积代替池化进行下采样,能够显著降低内存占用。

2.2 深度学习的优化策略

目前,基于深度学习的方法在脑肿瘤分割领域取得了较好的成绩,这归因于以下事实:深度学习方法通过堆叠若干卷积层来构造深度CNN,将图像与内核进行卷积以形成鲁棒性和自适应能力更强的分割模型。

尽管这些基于深度学习的方法取得了一定的进展,但在实践中仍然存在以下问题:

1) 脑部MRI图像难以获取,并且相对于整个MRI图像,脑肿瘤区域较小,背景区域明显多于目标对象,这种类别不平衡会加大分割难度;

2) 基本的网络结构限制了图像信息的利用和特征提取的能力,导致训练模型的分割精度提升有限;

3) 随着网络深度加深,学习效率会变低,分割准确率无法有效提升。

针对上述问题,将已提出的相应改进方法进行了归纳。在图像数据方面,利用如翻转、缩放、裁剪等数据增强技术(Krell等,2018Mok和Chung,2018)增加训练数据量,提高模型的泛化能力;结合脑部解剖结构中肿瘤区域包含关系,引入级联框架(Dvořák和Menze,2015Casamitjana等,2017Lachinov等,2018),分别实现全肿瘤、肿瘤核心和增强肿瘤的分割;使用改进的损失函数(Jesson和Arbel,2017Zhang等,2017)处理图像类别不平衡问题。在网络结构方面,采用多尺度、多通道策略(Havaei等,2015;Zhao和Jia,2016),充分利用图像特征信息;下采样过程中,用卷积操作代替池化(Sherman,2018),在减少图像信息丢失的同时,能够进一步对图像特征进行学习;在多个卷积层之间,使用跳跃连接(He等,2016Huang等,2017)的方式,有效解决深层次网络的退化问题;在不同情况下,选用合适的标准化方式、激活函数以及丢失率,以达到更好的分割效果。

3 BraTs挑战和方法发展趋势

脑肿瘤分割BraTS(brain tumor segmentation)挑战是用来评估MRI脑肿瘤分割方法的比赛。BraTS挑战的数据集使用多个机构脑部术前MRI图像数据,专注于神经胶质瘤的分割。此外,为了确定该分割任务的临床相关性,BraTS挑战还通过结合放射学特征和机器学习算法综合预测患者总生存期。

脑肿瘤图像分割方法日益增多,但脑部成像方式、肿瘤类型和疾病状态不同,导致数据集差异较大,以至于很难比较方法的优劣。为此,BraTS提供了一个大型数据集,并已进行配准和颅骨分离。该数据集包括充足的由多个机构常规临床获得的HGG和LGG的术前多模态(T1、T1C、T2和FLAIR)MRI图像,分别作为这项挑战的训练、验证和测试数据,并附有描述相关肿瘤区域的标签。数据标签包括坏死、水肿、非增强性肿瘤和增强肿瘤,由1到4位评估者按照相同注释协议手动分割,并得到了有经验的神经放射科医师的批准。

这项挑战中,涌现了许多成绩优异的分割方法,表 3为比赛分割结果排名前3的分割方法和Dice分数。

表 3 BraTS挑战中前3名方法总结
Table 3 Summary of the top three methods in the BraTS challenge

下载CSV
文献 方法 分割性能(Dice)
全肿瘤 肿瘤核心 增强肿瘤
Kamnitsas等人(2017a) 集成多模型和多种架构 0.886 0.785 0.729
Wang等人(2017) 级联各向异性卷积神经网络 0.873 9 0.774 8 0.783 1
Isensee等人(2018) U-Net变体,多类Dice损失函数,数据增强 0.858 0.775 0.647

本文通过学习和整理BraTS挑战所用方法,总结脑肿瘤分割方法发展趋势:

1) 由于MRI成像模态的多样化,充分利用各个模态图像信息能有效提高脑肿瘤分割精度,因此如何合理使用多模态图像将成为一个研究热点;

2) 基于深度学习的方法在脑肿瘤分割领域表现突出,成为一个热门的研究方向;

3) 机器学习算法的缺陷会导致脑肿瘤分割结果不准确,对原本方法进行有效改进或结合各种方法是目前一个流行趋势。

4 结语

基于MRI图像的脑肿瘤分割是生物医学工程和计算机应用技术的交叉领域。BraTS挑战提供的公共数据集为研究人员利用现有技术开发和客观评估其方法提供了公共的平台。

本文总结了现有脑部MRI图像脑肿瘤分割的主要方法,深入分析了各方法的优缺点,重点综述了基于深度学习最新方法的关键技术,并结合BraTS比赛情况展望了脑肿瘤分割领域的发展趋势。

尽管现有的方法取得了一定的进展,但脑肿瘤图像分割仍然存在一些挑战:

1) 脑肿瘤分割是一种异常检测问题,比其他基于模式识别的任务更具有挑战性;

2) 大多数方法能够精确分割HGG病例图像,但在LGG病例图像分割表现较差;

3) 与分割全肿瘤相比,这些方法对肿瘤核心和增强肿瘤的分割效果仍然不佳。

以上挑战将成为该领域学术界的研究焦点,基于深度学习的方法也将继续占据主流地位。

参考文献

  • Bauer S, Nolte L P and Reyes M. 2011a. Fully automatic segmentation of brain tumor images using support vector machine classification in combination with hierarchical conditional random field regularization//Proceedings of the 14th International Conference on Medical Image Computing and Computer-Assisted Intervention. Toronto: Springer: 354-361[DOI:10.1007/978-3-642-23626-6_44]
  • Bauer S, Nolte L P and Reyes M. 2011b. Segmentation of brain tumor images based on atlas-registration combined with a Markov-Random-Field lesion growth model//Proceedings of 2011 IEEE International Symposium on Biomedical Imaging: from Nano to Macro. Chicago: IEEE: 2018-2021[DOI:10.1109/ISBI.2011.5872808]
  • Bauer S, Wiest R, Nolte L P, Reyes M. 2013. A survey of MRI-based medical image analysis for brain tumor studies. Physics in Medicine and Biology, 58(13): R97-R129 [DOI:10.1088/0031-9155/58/13/R97]
  • Benson C C, Lajish V L and Rajamani K. 2015. Brain tumor extraction from MRI brain images using marker based watershed algorithm//Proceedings of 2015 International Conference on Advances in Computing, Communications and Informatics. Kochi, India: IEEE: 318-323[DOI:10.1109/ICACCI.2015.7275628]
  • Bezdek J C, Hall L O, Clarke L P. 1993. Review of MR image segmentation techniques using pattern recognition. Medical Physics, 20(4): 1033-1048 [DOI:10.1118/1.597000]
  • Bray F, Ferlay J, Soerjomataram I, Siegel R L, Torre L A, Jemal A. 2018. Global cancer statistics 2018:GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA:A Cancer Journal for Clinicians, 68(6): 394-424 [DOI:10.3322/caac.21492]
  • Cai H M, Verma R, Ou Y M, Lee S K, Melhem E R and Davatzikos C. 2007. Probabilistic segmentation of brain tumors based on multi-modality magnetic resonance images//Proceedings of the 4th IEEE International Symposium on Biomedical Imaging: From Nano to Macro. Arlington: IEEE: 600-603[DOI:10.1109/ISBI.2007.356923]
  • Capelle A S, Colot O, Fernandez-Maloigne C. 2004. Evidential segmentation scheme of multi-echo MR images for the detection of brain tumors using neighborhood information. Information Fusion, 5(3): 203-216 [DOI:10.1016/j.inffus.2003.10.001]
  • Casamitjana A, Catà M, Sánchez I, Combalia M and Vilaplana V. 2017. Cascaded V-Net using ROI masks for brain tumor segmentation//Proceedings of the 3rd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Quebec City, QC, Canada: Springer: 381-391[DOI:10.1007/978-3-319-75238-9_33]
  • Chen H, Dou Q, Yu L Q and Heng P A. 2016. VoxResNet: deep voxelwise residual networks for volumetric brain segmentation[EB/OL].[2019-10-03].https://arxiv.org/pdf/1608.05895.pdf
  • Chen L L, Wu Y, DSouza A M, Abidin A Z, Wismüller A and Xu C L. 2018. MRI tumor segmentation with densely connected 3D CNN//Proceedings Volume 10574, Medical Imaging 2018: Image Processing. Houston: SPIE: 10574[DOI:10.1117/12.2293394]
  • Corso J J, Sharon E, Dube S, El-Saden S, Sinha U, Yuille A. 2008. Efficient multilevel brain tumor segmentation with integrated bayesian model classification. IEEE Transactions on Medical Imaging, 27(5): 629-640 [DOI:10.1109/TMI.2007.912817]
  • Dam E, Loog M and Letteboer M. 2004. Integrating automatic and interactive brain tumor segmentation//Proceedings of the 17th International Conference on Pattern Recognition. Cambridge: IEEE: 790-793[DOI:10.1109/ICPR.2004.1334647]
  • DeAngelis L M. 2001. Brain tumors. New England Journal of Medicine, 344(2): 114-123 [DOI:10.1056/NEJM200101113440207]
  • Deng W K, Xiao W, Deng H and Liu J G. 2010. MRI brain tumor segmentation with region growing method based on the gradients and variances along and inside of the boundary curve//Proceedings of the 3rd International Conference on Biomedical Engineering and Informatics. Yantai: IEEE: 393-396[DOI:10.1109/BMEI.2010.5639536]
  • Dong H, Yang G, Liu F D, Mo Y H and Guo Y K. 2017. Automatic brain tumor detection and segmentation using U-Net based fully convolutional networks//Proceedings of the 21st Annual Conference on Medical Image Understanding and Analysis. Edinburgh: Springer: 506-517[DOI:10.1007/978-3-319-60964-5_44]
  • Dvořák P and Menze B H. 2015. Structured prediction with convolutional neural networks for multimodal brain tumor segmentation//Proceeding of the Multimodal Brain Tumor Image Segmentation Challenge. Munich: MICCAI: 13-24
  • Farag A A, Ahmed M N, El-Baz A and Hassan H. 2005. Advanced segmentation techniques//Suri J S, Wilson D L and Laxminarayan S, eds. Handbook of Biomedical Image Analysis: Volume Ⅰ: Segmentation Models Part A. Boston: Springer: 479-533[DOI:10.1007/0-306-48551-6_9]
  • Gibbs P, Buckley D L, Blackband S J, Horsman A. 1996. Tumour volume determination from MR images by morphological segmentation. Physics in Medicine and Biology, 41(11): 2437-2446 [DOI:10.1088/0031-9155/41/11/014]
  • Gies V and Bernard T M. 2004. Statistical solution to watershed over-segmentation//Proceedings of 2004 International Conference on Image Processing. Singapore: IEEE: 1863-1866[DOI:10.1109/ICIP.2004.1421440]
  • Gordillo N, Montseny E, Sobrevilla P. 2013. State of the art survey on MRI brain tumor segmentation. Magnetic Resonance Imaging, 31(8): 1426-1438 [DOI:10.1016/j.mri.2013.05.002]
  • Havaei M, Davy A, Warde-Farley D, Biard A, Courville A, Bengio Y, Pal C, Jodoin P M, Larochelle H. 2017. Brain tumor segmentation with deep neural networks. Medical Image Analysis, 35: 18-31 [DOI:10.1016/j.media.2016.05.004]
  • He K M, Zhang X Y, Ren S Q and Sun J. 2016. Deep residual learning for image recognition//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE: 770-778[DOI:10.1109/CVPR.2016.90]
  • Ho S, Bullitt E and Gerig G. 2002. Level-set evolution with region competition: automatic 3-D segmentation of brain tumors//Proceedings of the 16th International Conference on Pattern Recognition. Washington: IEEE: 10532
  • Hou Z J. 2006. A review on MR image intensity inhomogeneity correction. International Journal of Biomedical Imaging, 2006: 49515 [DOI:10.1155/IJBI/2006/49515]
  • Huang G, Liu Z, van der Maaten L and Weinberger K Q. 2017. Densely connected convolutional networks//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE: 4700-4708[DOI:10.1109/CVPR.2017.243]
  • Iftekharuddin K M, Zheng J, Islam M A, Ogg R J. 2009. Fractal-based brain tumor detection in multimodal MRI. Applied Mathematics and Computation, 207(1): 23-41 [DOI:10.1016/j.amc.2007.10.063]
  • Isensee F, Kickingereder P, Wick W, Bendszus M and Maier-Hein K H. 2017. Brain tumor segmentation and radiomics survival prediction: contribution to the brats 2017 challenge//Proceedings of the 3rd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Quebec City: Springer: 287-297[DOI:10.1007/978-3-319-75238-9_25]
  • Işın A, Direkoǧlu C, Şah M. 2016. Review of MRI-based brain tumor image segmentation using deep learning methods. Procedia Computer Science, 102: 317-324 [DOI:10.1016/j.procs.2016.09.407]
  • Jesson A and Arbel T. 2017. Brain tumor segmentation using a 3D FCN with multi-scale loss//Proceedings of the 3rd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Quebec City: Springer: 392-402[DOI:10.1007/978-3-319-75238-9_34]
  • Kamnitsas K, Bai W, Ferrante E, McDonagh S, Sinclair M, Pawlowski N, Rajchl M, Lee M, Kainz B, Rueckert D and Glocker B. 2017a. Ensembles of multiple models and architectures for robust brain tumour segmentation//Proceedings of the 3rd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Quebec City: Springer: 450-462[DOI:10.1007/978-3-319-75238-9_38]
  • Kamnitsas K, Ledig C, Newcombe V F J, Simpson J P, Kane A D, Menon D K, Rueckert D, Glocker B. 2017b. Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation. Medical Image Analysis, 36: 61-78 [DOI:10.1016/j.media.2016.10.004]
  • Kannan S R. 2008. A new segmentation system for brain MR images based on fuzzy techniques. Applied Soft Computing, 8(4): 1599-1606 [DOI:10.1016/j.asoc.2007.10.025]
  • Krell M M, Seeland A and Kim S K. 2018. Data augmentation for brain-computer interfaces: analysis on event-related potentials data[EB/OL].[2019-10-03].https://arxiv.org/pdf/1801.02730.pdf
  • Lachinov D, Vasiliev E and Turlapov V. 2018. Glioma segmentation with cascaded UNet//Proceedings of the 4th International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Granada: Springer: 189-198[DOI:10.1007/978-3-030-11726-9_17]
  • Lefkovits L, Lefkovits S and Szilágyi L. 2016. Brain tumor segmentation with optimized random forest//Proceedings of the 2nd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Athens: Springer: 88-99[DOI:10.1007/978-3-319-55524-9_9]
  • Letteboer M M, Olsen O F, Dam E B, Willems P W A, Viergever M A, Niessen W J. 2004. Segmentation of tumors in magnetic resonance brain images using an interactive multiscale watershed algorithm. Academic Radiology, 11(10): 1125-1138 [DOI:10.1016/j.acra.2004.05.020]
  • Liang Z P, Lauterbur P C. 2000. Principles of Magnetic Resonance Imaging:A Signal Processing Perspective. New York: The Institute of Electrical and Electronics Engineers Press
  • Long J, Shelhamer E and Darrell T. 2015. Fully convolutional networks for semantic segmentation//Proceedings of 2015 IEEE Computer Vision and Pattern Recognition. Boston: IEEE: 3431-3440[DOI:10.1109/CVPR.2015.7298965]
  • Luo S H, Li R X and Ourselin S. 2003. A new deformable model using dynamic gradient vector flow and adaptive balloon forces//APRS Workshop on Digital Computing. Brisbane, Australia: [s.n.]: 9-14
  • Madabhushi A, Udupa J K. 2005. Interplay between intensity standardization and inhomogeneity correction in MR image processing. IEEE Transactions on Medical Imaging, 24(5): 561-576 [DOI:10.1109/TMI.2004.843256]
  • Malladi R, Sethian J A, Vemuri B C. 1995. Shape modeling with front propagation:a level set approach. IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(2): 158-175 [DOI:10.1109/34.368173]
  • McInerney T and Terzopoulos D. 1996. Deformable models in medical image analysis//Proceedings of Workshop on Mathematical Methods in Biomedical Image Analysis. San Francisco: IEEE: 171-180[DOI:10.1109/MMBIA.1996.534069]
  • Mok T C W and Chung A C S. 2018. Learning data augmentation for brain tumor segmentation with coarse-to-fine generative adversarial networks//Proceedings of the 4th International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Granada: Springer: 70-80[DOI:10.1007/978-3-030-11723-8_7]
  • Nie J X, Xue Z, Liu T M, Young G S, Setayesh K, Guo L, Wong S T C. 2009. Automated brain tumor segmentation using spatial accuracy-weighted hidden Markov Random Field. Computerized Medical Imaging and Graphics, 33(6): 431-441 [DOI:10.1016/j.compmedimag.2009.04.006]
  • Olabarriaga S D, Smeulders A W M. 2001. Interaction in the segmentation of medical images:a survey. Medical Image Analysis, 5(2): 127-142 [DOI:10.1016/S1361-8415(00)00041-4]
  • Papageorgiou E I, Spyridonos P P, Glotsos D T, Stylios C D, Ravazoula P, Nikiforidis G N, Groumpos P P. 2008. Brain tumor characterization using the soft computing technique of fuzzy cognitive maps. Applied Soft Computing, 8(1): 820-828 [DOI:10.1016/j.asoc.2007.06.006]
  • Pereira S, Pinto A, Alves V, Silva C A. 2016. Brain tumor segmentation using convolutional neural networks in MRI images. IEEE Transactions on Medical Imaging, 35(5): 1240-1251 [DOI:10.1109/TMI.2016.2538465]
  • Pham D L, Xu C Y, Prince J L. 2000. Current methods in medical image segmentation. Annual Review of Biomedical Engineering, 2: 315-337 [DOI:10.1146/annurev.bioeng.2.1.315]
  • Pinto A, Pereira S, Dinis H, Silva C A and Rasteiro D M L D. 2015. Random decision forests for automatic brain tumor segmentation on multi-modal MRI images//Proceedings of the 4th Portuguese Meeting on Bioengineering. Porto: IEEE: 1-5[DOI:10.1109/ENBENG.2015.7088842]
  • Prastawa M, Bullitt E, Ho S, Gerig G. 2004. A brain tumor segmentation framework based on outlier detection. Medical Image Analysis, 8(3): 275-283 [DOI:10.1016/j.media.2004.06.007]
  • Rao C H, Naganjaneyulu P V and Prasad K S. 2017. Brain tumor detection and segmentation using conditional random field//Proceedings of the 7th International Advance Computing Conference. Hyderabad: IEEE: 807-810[DOI:10.1109/IACC.2017.0166]
  • Ren L, Li Q, Guan X, Ma J. 2018. Three-dimensional segmentation of brain tumors in magnetic resonance imaging based on improved continuous max-flow. Laser & Optoelectronics Progress, 55(11): 215-223 (任璐, 李锵, 关欣, 马杰. 2018. 改进的连续型最大流算法脑肿瘤磁核共振成像三维分割. 激光与光电子学进展, 55(11): 215-223) [DOI:10.3788/LOP55.111011]
  • Rexilius J, Hahn H K, Klein J, Lentschig M G and Peitgen H O. 2007. Multispectral brain tumor segmentation based on histogram model adaptation//Proceedings Volume 6514, Medical Imaging 2007: Computer-Aided Diagnosis. San Diego: SPIE, 6514: 65140V[DOI:10.1117/12.709410]
  • Ronneberger O, Fischer P and Brox T. 2015. U-Net: convolutional networks for biomedical image segmentation//Proceedings of the 18th International Conference on Medical Image Computing and Computer-Assisted Intervention. Munich: Springer: 234-241[DOI:10.1007/978-3-319-24574-4_28]
  • Ruan S, Lebonvallet S, Merabet A and Constans J. 2007. Tumor segmentation from a multispectral MRI images by using support vector machine classification//Proceedings of the 4th IEEE International Symposium on Biomedical Imaging: from Nano to Macro. Arlington: IEEE: 1236-1239[DOI:10.1109/ISBI.2007.357082]
  • Ruan S, Zhang N, Liao Q M and Zhu Y M. 2011. Image fusion for following-up brain tumor evolution//Proceedings of 2011 IEEE International Symposium on Biomedical Imaging: from Nano to Macro. Chicago: IEEE: 281-284[DOI:10.1109/ISBI.2011.5872406]
  • Sato M, Lakare S, Wan M, Kaufman A and Nakajima M. 2000. A gradient magnitude based region growing algorithm for accurate segmentation//Proceedings 2000 International Conference on Image Processing. Vancouver: IEEE: 448-451[DOI:10.1109/ICIP.2000.899432]
  • Schwartzbaum J A, Fisher J L, Aldape K D, Wrensch M. 2006. Epidemiology and molecular pathology of glioma. Nature Clinical Practice Neurology, 2(9): 494-503 [DOI:10.1038/ncpneuro0289]
  • Shen T, Huang X L, Li H S, Kim E, Zhang S T and Huang J Z. 2011. A 3D Laplacian-driven parametric deformable model//Proceedings of 2011 IEEE International Conference on Computer Vision. Barcelona: IEEE: 279-286[DOI:10.1109/ICCV.2011.6126253]
  • Sherman R. 2018. A volumetric convolutional neural network for brain tumor segmentation[EB/OL].[2019-10-03].https://arxiv.org/pdf/1811.0265401.pdf
  • Shi D L, Li Q, Guan X. 2018. Brain tumor image segmentation algorithm based on convolution neural network and fuzzy inference system. Journal of Frontiers of Computer Science and Technology, 12(4): 608-617 (师冬丽, 李锵, 关欣. 2018. 结合卷积神经网络和模糊系统的脑肿瘤分割. 计算机科学与探索, 12(4): 608-617) [DOI:10.3778/j.issn.1673-9418.1704042]
  • Siegel R, Naishadham D, Jemal A. 2013. Cancer statistics, 2013. CA:A Cancer Journal for Clinicians, 63(1): 11-30 [DOI:10.3322/caac.21166]
  • Stadlbauer A, Moser E, Gruber S, Buslei R, Nimsky C, Fahlbusch R, Ganslandt O. 2004. Improved delineation of brain tumors:an automated method for segmentation based on pathologic changes of 1H-MRSI metabolites in gliomas. NeuroImage, 23(2): 454-461 [DOI:10.1016/j.neuroimage.2004.06.022]
  • Stawiaski J. 2017. A multiscale patch based convolutional network for brain tumor segmentation[EB/OL].[2019-10-03].https://arxiv.org/pdf/1710.0231601.pdf
  • Sung Y C, Han K S, Song C J, Noh S M and Park J W. 2000. Threshold estimation for region segmentation on MR image of brain having the partial volume artifact//Proceedings of the 5th International Conference on Signal Processing Proceedings. 16th World Computer Congress 2000. Beijing: IEEE: 1000-1009[DOI:10.1109/ICOSP.2000.891695]
  • Tek H and Kimia B B. 1995. Shock-based reaction-diffusion bubbles for image segmentation//Proceedings of the 1st International Conference on Computer Vision, Virtual Reality and Robotics in Medicine. Nice: Springer: 434-438[DOI:10.1007/978-3-540-49197-2_55]
  • Tong Y F, Li Q, Guan X. 2018. An improved multi-modal brain tumor segmentation hybrid algorithm. Journal of Signal Processing, 34(3): 340-346 (童云飞, 李锵, 关欣. 2018. 改进的多模式脑肿瘤图像混合分割算法. 信号处理, 34(3): 340-346) [DOI:10.16798/j.issn.1003-0530.2018.03.011]
  • Vapnik V. 2000. The Nature of Statistical Learning Theory. New York: Springer
  • Vovk U, Pernus F, Likar B. 2007. A review of methods for correction of intensity inhomogeneity in MRI. IEEE Transactions on Medical Imaging, 26(3): 405-421 [DOI:10.1109/TMI.2006.891486]
  • Wang G T, Li W Q, Ourselin S and Vercauteren T. 2017. Automatic brain tumor segmentation using cascaded anisotropic convolutional neural networks//Proceedings of the 3rd International Workshop on Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Quebec City: Springer: 178-190[DOI:10.1007/978-3-319-75238-9_16]
  • Węgliński T and Fabijańska A. 2011. Brain tumor segmentation from MRI data sets using region growing approach//Perspective Technologies and Methods in MEMS Design. Polyana: IEEE: 185-188
  • Wong K P. 2005. Medical image segmentation: methods and applications in functional imaging//Suri J S, Wilson D L and Laxminarayan S, eds. Handbook of Biomedical Image Analysis: Volume Ⅱ: Segmentation Models Part B. Boston: Springer: 111-182[DOI:10.1007/0-306-48606-7_3]
  • Wu W, Chen A Y C, Zhao L, Corso J J. 2014. Brain tumor detection and segmentation in a CRF (conditional random fields) framework with pixel-pairwise affinity and superpixel-level features. International Journal of Computer Assisted Radiology and Surgery, 9(2): 241-253 [DOI:10.1007/s11548-013-0922-7]
  • Xing B T. 2018. MR Brain Tumor Image Segmentation Algorithm Based on Fully Convolutional Neural Network. Tianjin: Tianjin University (邢波涛. 2018. 基于全卷积神经网络的MR脑肿瘤图像分割算法研究. 天津: 天津大学)
  • Zhang J C, Shen X L, Zhuo T Q and Zhou H. 2017. Brain tumor segmentation based on refined fully convolutional neural networks with a hierarchical dice loss[EB/OL].[2019-10-03].https://arxiv.org/pdf/1712.09093.pdf
  • Zhao L Y, Jia K B. 2016. Multiscale CNNs for brain tumor segmentation and diagnosis. Computational and Mathematical Methods in Medicine, 2016: 8356294 [DOI:10.1155/2016/8356294]
  • Zhao X M, Wu Y H, Song G D, Li Z Y, Zhang Y Z, Fan Y. 2018. A deep learning model integrating FCNNs and CRFs for brain tumor segmentation. Medical Image Analysis, 43: 98-111 [DOI:10.1016/j.media.2017.10.002]