【信息技术】【2014.04】脑肿瘤的多参数成像和磁共振图像纹理分析

本文为加拿大西安大略大学（作者：Harish A. Sharma）的博士论文，共154页。

本文论述了临床医师在脑肿瘤治疗过程中面临的一些挑战。多形性胶质母细胞瘤（GBM，四级）是原发性脑肿瘤最恶性的形式，治疗后复发是很常见的。无创成像是脑肿瘤治疗计划和监测的重要组成部分。不幸的是，GBM患者的肿瘤复发和放射损伤（RI）在随访常规磁共振成像（MRI）中表现相似，因此很难选择最合适的治疗方案。继发性脑肿瘤的脑转移在全身性癌症患者中很常见。传统的磁共振成像也很难区分恶性肿瘤和转移瘤，但对指导手术治疗和放射治疗至关重要。因此，本论文的总体目标是发展并提高脑肿瘤检测的影像学方法。

第一个目标是利用常规MRI信号强度（T2加权（T2w）和液体衰减反演恢复（FLAIR））以及扩散张量成像参数（分数各向异性（FA）和径向扩散系数（RD））的多参数特征来鉴别GBM肿瘤复发和RI。在RI区域，FA和RD之间以及T2w和FLAIR信号强度之间存在显著相关性，而在肿瘤区域未观察到这种相关性。这些相关性可能有助于区分肿瘤复发和RI。

第二个目标是区分GBM和最常见的两种脑肿瘤转移（MET）。两者在常规磁共振成像中表现出相似的放射学表现，但需要不同的治疗策略。本研究的目的是探讨常规磁共振图像（对比增强T1加权图像）基于纹理的图像分析是否能够提供定量信息，以用于区分GBM和MET。结果表明，标准差的一阶纹理特征和熵、惰性、均匀性和能量的二阶纹理特征在两组之间存在显著差异。接收机工作特性（ROC）曲线分析表明，结合一阶和二阶特征能够提高GBM和MET的预测精度。

最后，螺旋体断层治疗（HT）是一种放射治疗技术，允许多发性脑转移同时进行放射外科治疗（SIB）和全脑放射治疗（WBRT）。然而，有些患者的肿瘤可能对HT型WBRT+SIB无反应。本研究的目的是探讨在HT型WBRT+SIB治疗前进行的磁共振成像（MRI）扫描对肿瘤大小和外观的定量测量是否可用于区分有反应和无反应的患者组。我们的研究结果表明，较小的病变可能对这种放射治疗有更好的反应。对于响应预测而言，外观测量比尺寸测量提供的附加信息有限。单独进行的边缘增强和核心坏死的定量测量并没有提供更多的预测价值。

综上所述，基于相关性的肿瘤与RI鉴别方法、基于定量纹理特征的GBM与MET鉴别方法，以及正确选择对HT型放射治疗有反应的患者，都可用于更好地规划患者治疗方案。

This thesis addresses some of thechallenges that clinicians face in the course of treatment of brain tumors.Glioblastoma multiforme (GBM, grade IV) is the most malignant form of primarybrain tumor and recurrence following treatment is common. Non-invasive imagingis an important component of brain tumor treatment planning and monitoring.Unfortunately, tumor recurrence and radiation injury (RI) in patients with GBMhave similar appearances on follow-up conventional magnetic resonance imaging(MRI), making it difficult to choose the most appropriate treatment plan. Brainmetastases which are secondary brain tumors are common in patients withsystemic cancer. Differentiating between GBM and metastatic tumor is alsodifficult with conventional MRI, but is essential for guiding surgical andradiotherapy treatment. Therefore, the overall goal of this thesis is todevelop imaging methods that improve brain tumor detection. The first objectivewas to develop a method to discriminate between GBM tumor recurrences and RIusing a multiparametric characterization of the tissue incorporatingconventional MRI signal intensities (T2-weighted (T2w) and fluid attenuatedinversion recovery (FLAIR)) and diffusion tensor imaging parameters (fractionalanisotropy (FA) and radial diffusivity (RD)). In the RI region there weresignificant correlations between FA and RD as well as between T2w and FLAIRsignal intensities. No such correlations were observed in the tumor region.These correlations may aid in differentiating between tumor recurrence and RI.The second objective was to differentiate between GBM and metastasis (MET); thetwo most common types of brain tumors. Both exhibit similar radiologicappearance on routine MR imaging but require different treatment strategies.The goal of this study was to investigate whether texture based image analysisof routine MR images (contrastenhanced T1-weighted images) would providequantitative information that could be used to differentiate between GBM andMET. Our results demonstrate that first-order texture feature of standarddeviation and second-order texture features of entropy, inertia, homogeneity,and energy show significant differences between the two groups. Receiveroperating characteristic (ROC) curve analysis showed that combining first- andsecondorder features increased the predictive accuracy in differentiatingbetween GBM and MET. Finally, helical tomotherapy (HT) is a type of radiationdelivery technique that allows for a radiosurgery-type simultaneous infieldboost (SIB) of multiple brain metastases, synchronously with whole brainradiation therapy (WBRT). However, some patients’ tumors may not respond to HTtype WBRT+SIB. The goal of our study was to investigate whether quantitativemeasurements of tumor size and appearance on magnetic resonance imaging (MRI)scans acquired prior to HT type WBRT+SIB treatment could be used todifferentiate responder and non-responder patient groups. Our resultsdemonstrated that smaller size lesions may respond better to this type ofradiation therapy. Measures of appearance provided limited added value overmeasures of size for response prediction. Quantitative measurements of rimenhancement and core necrosis performed separately did not provide additionalpredictive value. In summary, our correlation based method for differentiatingtumor from RI, differentiating GBM and MET using quantitative texture features,and correctly selecting patients who will respond to HT type radiationtreatment may be used to better plan patient treatment.

1 引言
2 胶质母细胞瘤灌注CT高渗透区的形态MRI参数和扩散张量参数分析：从放射诱导的坏死区域识别肿瘤复发
3 利用一阶和二阶磁共振图像纹理区分胶质母细胞瘤和转移瘤
4 利用定量大小和外观特征预测病灶内同时增强的全脑螺旋断层扫描对转移性脑癌的反应
5 结论与未来工作展望

下载英文原文地址：

http://page2.dfpan.com/fs/bl8cbj421271e299160/

更多精彩文章请关注微信号：