【作者】 李卫娜；

【导师】 郑小林；

【作者基本信息】 重庆大学 ， 生物医学工程， 2019， 博士

【摘要】 肌萎缩侧索硬化(Amyotrophic Lateral Sclerosis,ALS)是一种同时累及上、下运动神经元的进行性神经退行性疾病,平均生存期3-5年。仅有5-10%的ALS患者为家族遗传,90%以上的病例中,ALS的发病都很突然,称为“散发性ALS(sporadic ALS,sALS)”,其致病原因尚不清楚。由于没有明确的临床诊断检测方案,目前sALS的诊断主要依赖于有经验的医生通过详细病史询问和体格检查,寻找上、下运动神经元共同受累及进行性加重的证据。然而不幸的是,对于这种快速进展的疾病,从症状出现到确诊平均有1年的延迟,严重妨碍了sALS的早期干预。建立客观的、可靠的sALS诊断生物标记物将代表sALS临床研究的重大进展。现代磁共振技术对sALS的研究已经取得了一定的进展,但研究内容较分散,没有对sALS脑结构特征和功能特征的系统研究,有些研究结果甚至相悖,且大多数研究都是发现结构或功能模式的改变,而没有形成影像学标记。基于此,本文利用现代磁共振技术系统的研究了sALS患者脑结构特征和脑功能特征,并基于这些脑结构、功能特征和支持向量机技术训练了系列模式分类器,以期为sALS的诊断提供客观的影像标记和个体化的诊断工具。具体如下:1.sALS的脑结构特征研究:本文利用基于体素的形态学测量方法和基于表面的形态学测量方法系统的研究了sALS患者的灰质体积、皮层厚度、脑沟深度、皮层表面积、平均曲率等特征,并分别基于灰质体积和皮层厚度构建了sALS患者的全脑结构网络,同时利用基于图论的脑网络分析方法研究了sALS患者在脑结构网络全局特征属性和局部特征属性方面的改变;最后探索了这些显著改变的脑结构特征与临床变量之间的相关性。研究结果发现,sALS患者初级运动区的灰质体积和皮层厚度较健康对照组均存在显著降低,符合sALS为运动神经元病的特征;此外,sALS患者顶下小叶的灰质体积、颞上回和额上回的皮层厚度较健康对照组也存在显著下降,表明sALS对皮层的损害并不局限在运动区,而是累及额叶、颞叶、顶叶等多个脑区;sALS的全脑结构网络分析显示,无论是灰质体积网络还是皮层厚度网络,同健康对照组一样,其脑结构网络都存在小世界属性,但随着网络连接密度的增加,其皮层厚度网络的小世界属性可能会丧失;此外,sALS患者的灰质体积网络和皮层厚度网络均显示网络的局部效率显著降低,提示sALS患者脑结构网络局部信息传递能力的减弱。相关分析发现,sALS患者初级运动区的灰质体积与其疾病病程呈显著负相关,初级运动区的皮层厚度与其ALSFRS_R评分呈显著正相关,使初级运动区成为诊断sALS的潜在结构影像学标记。2.sALS的脑功能特征研究:本文利用静息态fMRI影像数据并回顾文献中新近静息态fMRI特征提取方法,分别从脑功能的局部活动特征(包括低频幅值)、局部同步或连接特征(包括局部神经活动一致性、基于体素-镜像同伦的功能连接、短程功能连接密度、长程功能连接密度)及全脑连接特征(包括全脑功能连接强度、基于体素的度中心性)的角度系统的研究了sALS患者的脑功能特征,并构建了sALS患者的全脑功能连接网络,利用图论的分析方法研究其脑功能连接网络全局特征属性和局部特征属性的改变;最后探索这些显著改变的脑功能特征与临床变量间的相关性。研究结果发现,sALS患者与健康对照组脑功能特征差异最多的脑区为初级运动区和内侧和旁扣带回,其他还分布在额叶、颞叶、顶叶、枕叶和非皮层运动区(如壳核和小脑)等多个脑区;结构是功能的基础,sALS患者与正常对照组结构特征存在显著差异的脑区功能特征也存在显著差异,然而一些尚未显现结构特征差异的脑区,如内侧和旁扣带回,也已经表现出了明显的功能特征差异,提示我们sALS的早期诊断不仅要关注运动脑区,非运动脑区的功能特征也是一个重要指标。两组被试的全脑功能网络均具有小世界属性,但无论其全局特征属性还是局部特征属性均无显著差异。sALS患者初级运动区、内侧和旁扣带回等脑区功能特征的改变为sALS的诊断提供潜在功能影像学标记。3.基于支持向量机的sALS的个体化诊断:基于前面所得到的具有显著组间差异的脑结构特征、脑功能特征,利用支持向量机技术,采用“重复嵌套交叉验证”的模型训练和评估方法,本文构建了9个基于单个脑结构特征或功能特征的单模态模式分类器,11组基于多个脑结构或脑功能特征的多模态模式分类器和1个基于显著差异脑结构或脑功能特征脑区的多模态感兴趣区模式分类器。分类效果最好的单模态模式分类器为皮层厚度分类器(准确度:83.1%,敏感度:82.4%,特异度:83.9%,ROC曲线下面积:0.9);分类效果最好的多模态分类器为由皮层厚度分类器和VMHC分类器组成的多模态分类器(准确度:86.2%,敏感度:91.2%,特异度:80.6%,ROC曲线下面积:0.93),分类效果优于所有的单模态分类器;多模态ROI分类器的分类效果在所有训练的分类器中效果最优(准确度:98.5%,敏感度:100%,特异度:96.8%,ROC曲线下面积:0.99),但该分类器有过拟合风险。初级运动区基本上是所有单模态分类器对分类结果贡献显著(或贡献率大于95%)的区域,可以视作sALS诊断的标记性脑区。本文证明了利用sALS患者的脑结构或脑功能影像特征建立个体化水平的sALS诊断方法的可行性,为临床诊断sALS提供客观依据。综上所述,本文利用现代磁共振技术和影像数据分析方法系统的研究了sALS患者脑结构特征和脑功能特征的改变,并利用这些脑结构或脑功能特征提出了新的个体化水平诊断sALS的方法。本文主要有以下三项意义:(1)通过对sALS脑结构特征、功能特征及个体化诊断的研究发现初级运动区可以视作sALS诊断的标记性脑区,初级运动区的脑结构特征和功能特征应是诊断sALS的关注重点,此外,非运动脑区(如内侧和旁扣带回)的脑功能特征也是早期诊断sALS的重要指标;(2)通过横向对比基于单个脑结构特征或功能特征的单模态模式分类器的分类效果,发现皮层厚度分类器的分类效果最优,提示皮层厚度可能是诊断sALS最具代表性的影像特征;(3)利用支持向量机技术,建立了基于脑结构和功能特征的sALS个体化诊断模式分类器,其中皮层厚度分类器及基于皮层厚度特征和VMHC特征的多模态分类器取得了较理想的分类效果,为临床sALS的客观诊断提供帮助,提示基于影像特征和机器学习的sALS个体化诊断策略未来有望成为替代现行EI Escorial标准的诊断技术。更多还原

【Abstract】 Amyotrophic lateral sclerosis(ALS)is a progressive neurodegenerative disease involving both upper motor neuron and lower motor neuron,with a mean survival time between 3 and 5 years.About 5 to 10 percent of ALS is familial;the other 90 to 95 percent of ALS is sporadic,meaning it occurs without a family history(sporadic ALS,sALS),and the core etiology of the disease remains elusive.As there is no definitive diagnostic test for sALS,the diagnosis of sALS mainly relies on experienced doctors,through detailed medical history inquiry and physical examination,to search for evidence of the joint involvement and progressive aggravation of upper and lower motor neurons.For this rapidly progressive disease,there is an average delay of one year from symptom appearance to diagnosis,which seriously impedes the early intervention of sALS.The establishment of objective and reliable biomarkers for upper motor neuron degeneration will represent a significant advance in clinical studies of sALS.Modern magnetic resonance imaging(MRI)technology has made some progress in the research of sALS,but the research content is scattered,and there is no systematic study on the brain structural and functional features of the sALS.Some of the research results are even contrary,and most of the studies just find the changes of the brain structure or functional mode,without forming imaging markers.So,this dissertation studies the features of brain structure and brain function in patients with sALS characteristics systematically based on modern system of MRI technology.Besides,this dissertation also trains a series of pattern recognition classifiers based on support vector machine(SVM)technology,so as to provide objective imaging markers for the diagnosis of sALS and the individual diagnostic tool.The main work of this dissertation is summarized as below:1.Research on structural features of patients with sALS: this dissertation studied features of gray matter volume,cortical thickness,sulcal depth,surface area and average curvature of patients with sALS based on both voxel-based morphometry(VBM)method and surface-based morphometry(SBM)method.The dissertation also built the whole brain structural network based on gray matter volume and cortical thickness respectively to study the alterations of global and local characters of the brain network based on graph theroy.Finally,correlation between these significantly altered brain structural characters and clinical variables was explored.The results showed that the volume and cortical thickness of gray matter in the primary motor area of patients with sALS were significantly lower than those in the healthy control group,which was consistent with the characteristics of sALS as motor neuron disease.In addition,the gray matter volume in subparietal lobule,the cortical thickness in superior temporal gyrus and superior frontal gyrus also decreased significantly compared with the healthy control group,indicating that the damage of sALS to the cortex was not limited to the motor area,but involved multiple brain areas such as the frontal lobe,temporal lobe and parietal lobe.As the healthy control group,sALS patients showed that both the gray matter volume network and the cortical thickness network followed a small-world organization.But with the increase of network density,small-world organization of cortical thickness network would be lost.In addition,both the gray matter volume network and the cortical thickness network in patients with sALS showed a significant decrease in the local efficiency of the network,suggesting that the local information transmission ability of the brain structure network in patients with sALS was weakened.Correlation analysis showed that the change of gray matter volume in the primary motor area was significantly negatively correlated with the disease duration of sALS,and the cortical thickness in the primary motor area was significantly positively correlated with the ALSFRS_R score,making the primary motor area a potential structural imaging marker for the diagnosis of sALS.2.Research on functional features of patients with sALS: this dissertation studied functional features of sALS from local character(including amplitude of low frequency fluctuation),local functional connection character(including regional homogeneity,voxel-mirrored homotopic connectivity,short-range and long-range functional connectivity density)and whole brain functional conneciton character(including degree centrality and brain functional connectivity)systematically based on the review of newly resting state fMRI feature extraction method.Additionally,whole brain functional connection network was constructed and changes of both global network measures and local network measures were studied.Finally,the correlation between these significantly altered brain functional characteristics and clinical variables was explored.The results showed that besides the primary motor area,medial cingulate gyrus was also the brain area with the most differences in functional features between sALS patients and the healthy control group.The involved brain areas also included parietal lobe,occipital lobe,marginal lobe,subcortical nucleus and cerebellum.Structure is the basis of brain function.Brain regions with significant difference in brain structural features also had significant difference in brain functional features.Moreover,some brain regions(i.e.median cingulate and paracingulate gyri)showed altered functional features even before they showed altered structural features,which suggested that early diagnosis of sALS should pay attention not only to the motor brain area,but also to the functional characteristics of the non-motor brain area.The whole brain functional connection network of sALS patients also show small-world characterstics,but there was no significant difference in global or regional network measures between sALS patients and healthy control group.The change of functional features in primary motor area and other brain areas of sALS patients provided potential functional imaging markers for the diagnosis of sALS.3.Individual diagnosis of sALS based on support vector machine(SVM): this dissertation trained series of classifiers to diagnose sALS individually based on brain structural and functional features mentioned previously and support vector machine(SVM)model.A repeated nested cross-validation method was used to strictly separate the train process from the evaluation of the classification model’s generalization capacity.The classification models included 9 single-mode classifiers based on single brain structural or functional feature,11 multi-mode classifiers based on several brain structural or functional features and 1 multi-mode ROI(regions of interest)classifier based on multi-mode brain regions with significantly different brain structural or functional features.Single-mode classifier with the best classification performance was the cortical thickness classifier(accuracy: 83.1%,sensitivity: 82.4%,specificity: 83.9%,the area under the ROC curve: 0.9).Multi-mode classifier with the best classification performance was fused by cortical thickness classifier and VMHC classifier(accuracy: 86.2%,sensitivity: 91.2%,specificity: 80.6%,the area under the ROC curve: 0.93).And the ROI classifier had the best classification performance of all trained classifiers with accuracy of 98.5%,sensitivity of 100%,specificity of 96.8% and the area under the ROC curve of 0.99,but the ROI classifier had a risk of overfitting.Primary motor area was the brain region that contributed significantly(or contribution rate greater than 95%)for the classification results in nearly all singlemode classifiers.It makes this brain region a potential bio-marker for the diagnosis of sALS.This part of work proved the feasibility of establishing individualized diagnosis tools for sALS by using the brain mapping features,and provide objective basis for clinical diagnosis of sALS.In general,this dissertation systematically studied the changes of brain structural and functional features in patients with sALS based on modern magnetic resonance imaging technology and data analysis methods,and established a new individual level diagnosis tool of sALS by using these features.Through the work of this dissertation,the following preliminary conclusions could be obtained:(1)structural and functional features of the primary motor area should be the focus in the diagnosis of sALS,and the functional features of the non-motor brain areas(e.g.median cingulate and paracingulate gyri)were also important indicators;(2)cortical thickness might be the most representative feature in the diagnosis of sALS by comparing the classification effect of single mode classifier;(3)With the favorable results,this dissertation suggested that the individualized sALS diagnosis strategy based on image characteristics and machine learning might achieve better results than the current EI Escorial standard in the nearly future.更多还原