LightGBM算法在阿尔茨海默症结构磁共振成像分类中的应用

为更好地利用计算机技术分析阿尔茨海默症(AD)患者的大脑脑区变化,并对AD进行辅助诊断,本研究选择来自AD神经影像数据库的116名AD患者、116名轻度认知障碍患者和117名正常对照者的脑部结构磁共振成像,并利用spm软件对3组数据进行预处理和统计学相关性分析,得到差异性脑区。然后使用IBASPM软件提取病灶脑区体积作为特征样本。最后利用LightGBM算法对特征向量进行分类,并与支持向量机和XGBoost算法作对比实验。实验结果显示,利用LightGBM算法对病灶脑区的体积进行分类,准确率可达到83%。在这3种分类算法中,LightGBM更具有优势,分类结果更准确,可见,利用LightGBM算法可以有效地辅助医疗人员对AD进行早期诊断。     

颞叶癫痫患者脑白质纤维束追踪空间统计分析与自动识别

为了定位颞叶癫痫(TLE)患者脑白质微结构发生异常的重要脑区,本文设立了正常对照组(NC)与TLE组两组人群,采集了50位受试者(其中NC组28人,TLE组22人)的脑部弥散张量成像(DTI)影像,分别计算其部分各向异性(FA)、平均扩散率(MD)、扩散系数(AD)、径向扩散系数(RD)等参数,并采用纤维束追踪空间统计方法(TBSS),获取组间差异的脑区,然后利用支持向量机(SVM),对NC组与TLE组进行分类,并与支持向量机-递归特征消除法(SVM-RFE)进行比较,最后对重要脑区及其分布进行分析与讨论。实验结果表明,TLE患者的FA值存在明显降低的脑区主要有胼胝体、上纵束、放射冠、外囊、内囊、下额枕束、钩束、矢状层等,基本呈双侧分布,其中大部分脑区的MD、RD值明显增高,AD值虽有增高,但差异无统计学意义。支持向量机-纤维束追踪空间统计法(SVM-TBSS)利用FA、MD、RD进行分类的准确率分别为82%、76%、76%,特征融合后分类准确率为80%;SVM-RFE利用FA、MD、RD进行分类准确率分别为90%、90%和92%,特征融合后分类准确率达到100%,SVM-RFE分类性能明显优于SVM-TBSS,对分类有重要影响的特征主要分布于联络纤维和连合纤维脑区。研究结果表明,DTI参数能有效地反映TLE患者的脑白质纤维异常改变,可用于阐明其病理机制、定位病灶及实现自动诊断。     

基于3D卷积神经网络-感兴趣区域的阿尔茨海默症辅助诊断模型

磁共振(magnetic resonance imaging,MRI)图像的预测分类对早期阿尔茨海默症(Alzheimer′s disease,AD)的诊断非常重要。轻度认知障碍(mild cognitive impairment,MCI)作为AD的一种早期阶段,在诊断时存在大脑脑区萎缩区域不明确,诊断准确率偏低等问题。本研究提出一种基于感兴趣区域(regions of interest,ROI)的3D卷积神经网络(convolutional neural network, CNN)模型来解决AD分类准确率偏低等问题,进而实现对AD的计算机辅助诊断。实验数据均来自ADNI数据库,实验结果表明,基于ROI的3D CNN的AD辅助诊断模型在分类AD vs正常对照(normal control,NC)、MCI转化AD(MCI converted to AD,MCIc) vs NC和MCI未被转化AD(MCI not converted to AD,MCInc) vs MCIc的5折交叉验证平均准确率分别为85.2%、83.9%、68.5%。相比于传统的主成分分析+支持向量机方法和单纯的切片集成方法,本研究方法在AD辅助诊断中取得了更好的分类效果和泛化能力,还可为其他脑疾病诊断提供新思路。     

KPCA和Adaboost算法在阿尔茨海默症功能磁共振影像分类中的 应用

本研究的目的在于使用机器学习方法,对脑部功能磁共振成像数据进行分析与特征提取,完成对阿尔茨海默症 （AD）的辅助诊断与分析。首先对数据进行预处理与去除协变量,并从大脑全局特征出发,根据现有的自动解剖标记模 板,把每个被试的大脑分为116个脑区,通过提取每个脑区的时间序列,构建全脑功能连接矩阵,然后使用核主成分分析 法进行特征提取,最后用Adaboost算法进行分类。在对34名AD患者、35名轻度认知障碍患者和35名正常对照组的功能 磁共振成像数据进行的实验结果表明,利用静息态功能磁共振成像,同时结合机器学习的方法,能够有效地实现AD的正 确分类,准确率可以达到96%,该结果可以为AD患者的临床辅助诊断提供有效的判断依据。     

基于多模态磁共振影像的首发未用药精神分裂症自动分类研究

慢性精神分裂症患者大脑的结构和功能异常已经被广泛报道,但是首发未用药精神分裂症患者和正常人的相关研究较少。本研究采集了44名首发未用药精神分裂症患者和56名正常人的结构和静息态功能磁共振图像,基于自动解剖标签模板提取了90个感兴趣区域的灰质体积、局部一致性、低频振荡振幅和度中心度作为特征,并将这些特征作为输入,用基于递归特征消除的支持向量机对首发未用药精神分裂症患者和正常人进行分类。结果表明,局部一致性和低频振荡振幅的组合为最佳分类特征,分类准确率达到96.97%,并且分类权重最大的脑区主要位于额叶。研究结果有利于加深对精神分裂症神经病理机制的了解,有助于开发出用于临床辅助诊断的生物学标记物。     

阿尔茨海默病患者扣带部位磁共振图像纹理特征及性别特点

目的 分析阿尔茨海默（Alzheimer＇s disease,AD）、轻度认知障碍（mild cognitive impairment,MCI）和正常对照者（normal controls,NC）MR图像扣带纹理特征,并按性别进行分组分析,探索纹理特征在疾病早期诊断上的应用.方法 利用灰度共生矩阵和游程长矩阵对52例MR图像（AD14例、MCI20例、NC18例）进行纹理分析,测试组间参数是否显著不同,并用支持向量机方法对在组间具有显著性差异的纹理特征进行分类识别.结果 与NC组相比,AD和MCI组的能量、游程长不均匀度等纹理特征在前扣带和后扣带均存在显著性差异.按性别分组的实验结果显示,除AD组和NC组间存在显著不同外,男性后扣带在MCI与NC及MCI与AD间均存在显著性差异（P〈0.05）,女性前扣带在MCI和NC组间存在显著性差异（P〈0.05）.分类识别结果显示,男性后扣带分类正确率最高,为90%.结论 MR图像纹理特征可以反映扣带病理变化,不同性别MR图像表现不同,有可能为AD的早期诊断提供帮助.     

基于sMRI的阿尔茨海默症分类影响因素研究

本研究提出基于三类解剖特征的SVM建模方法,探索样本、特征及算法选择三个因素,对阿尔茨海默症(AD)及其前驱阶段分类的重要性。该方法以三维重构s MRI后不同大脑区域的灰质体积、皮层表面积及其平均厚度三类特征作为SVM模型的输入参数,并采用十折交叉验证方法对AD患者、轻度认知损害患者和健康者进行分类识别,并与其他文献结果进行比较分析。实验结果表明,为了达到更高的分类准确率,选择合适的样本和特征,比选择算法更重要。此结论为未来AD的计算机辅助诊断研究工作提供了有益的指导。     

基于非线性高阶特征和超图卷积神经网络的阿尔茨海默症分类

阿尔茨海默症（AD）是一种不可逆转的大脑神经退化性疾病，会损害患者记忆力和认知能力。因此，AD诊断具有重要意义。大脑感兴趣区域（ROI）之间往往是多个区域以非线性的方式协同交互，充分利用此类非线性高阶交互特征有助于提高AD诊断分类的准确性。为此，提出基于非线性高阶特征提取和三维超图神经网络相结合的AD计算机辅助诊断框架。首先针对ROI数据使用基于径向基函数核的支持向量机回归模型训练出基估计器，再通过基于基估计器的递归特征消除算法提取功能性磁共振成像（fMRI）数据中的非线性高阶特征，进而将特征构造成超图，最后基于fMRI数据的四维时空特性搭建超图卷积神经网络模型来进行分类。阿尔茨海默症神经影像倡议（ADNI）数据库上的实验结果表明，所提框架在AD/正常对照（NC）分类任务上的效果相较于Hyper Graph Convolutional Network(HyperGCN)框架提高了8%，相较于传统二维线性特征提取方法提高了12%。综上，本文框架在AD分类效果上较主流深度学习方法有所提升，可为AD计算机辅助诊断提供有效依据。     

阿尔茨海默病患者松果体MRI研究

目的　观察阿尔茨海默病（Alzheimer’s disease,AD)发病进程中松果体的MRI形态学改变。 方法　依据AD发病进程,分别采集NC组24例、MCI组18例、轻度AD组16例、中-重度AD组20例受试者大脑MRI扫描图像,测量松果体矢径、横径、高、体积,并分析组间各测量值变化趋势。 结果　松果体标化高MCI期、轻度AD期较NC组增大（P＜0.05）,中-重度AD期标化高较MCI期、轻度AD缩小（P＜0.05）;松果体标化矢径、标化横径、标化体积MCI期、轻度AD、中-重度AD与NC组比无明显变化（P＞0.05）,但中-重度AD较MCI期组缩小（P＜0.05）。 结论　在AD发病进程中松果体高先增大再萎缩,松果体高对MCI期、轻度AD患者的诊断有较大价值。     

基于随机森林算法的阿尔茨海默症医学影像分类

为实现阿尔茨海默症(AD)的医学影像分类,辅助医生对患者的病情进行准确判断,本研究对采集的34名AD患者、35名轻度认知障碍患者和35名正常对照组成员的功能磁共振影像进行特征提取和分类,具体思路包括:首先利用皮尔逊相关系数计算脑区之间的功能连接,然后采用随机森林算法对被试不同脑区之间的功能连接进行重要性度量及特征选择,最后使用支持向量机分类器进行分类,利用十倍交叉验证估算分类准确率。实验结果显示,随机森林算法可以对功能连接特征进行有效分析,同时得到AD发病过程的异常脑区,基于随机森林和SVM建立的分类模型对AD、轻度认知障碍的识别具有较好的效果,分类准确率可达90.68%,相关结论可以为AD的早期临床诊断提供客观参照。 【关键词】阿尔茨海默症;功能磁共振成像;随机森林;特征选择     

基于KPCA算法的阿尔茨海默症辅助诊断

阿尔茨海默症（AD）是一种起病隐匿、进行性发展的神经系统退行性疾病,利用磁共振成像和计算机技术对AD患者的辅助诊断是目前不断探索的新课题。本研究先对磁共振图像进行预处理和相关性分析,然后利用核主成分分析法（KPCA）对脑灰质图像进行特征提取,结合Adaboost算法进行分类,并与主成分分析法（PCA）进行对比试验。通过对AD神经影像学计划数据库中的116名AD患者、116名轻度认知障碍患者,以及117名正常对照的脑部功能磁共振成像进行的研究表明,利用机器学习能够很有效地辅助诊断AD脑部疾病,KPCA算法对图像进行特征提取比PCA 算法更加充分完备,分类结果更加精确,能够获得更好的AD辅助诊断结果。     

Tissue segmentation-assisted analysis of fMRI for human motor response: an approach combining artificial neural network and fuzzy C means

The authors have developed an automated algorithm for segmentation of magnetic resonance images (MRI) of the human brain. They investigated the quantitative analysis of tissue-specific human motor response through an approach combining gradient echo functional MRI and automated segmentation analysis. Fifteen healthy volunteers, placed in a 1.5 T clinical MR imager, performed a self-paced finger opposition throughout the activation periods. T1-weighted images (WI), T2WI, and proton density WI were acquired for segmentation analysis. Single-slice axial T2* fast low-angle shot (FLASH) images were obtained during the functional study. Pixelwise cross-correlation analysis was performed to obtain an activation map. A cascaded algorithm, combining Kohonen feature maps and fuzzy C means, was applied for segmentation. After processing, masks for gray matter, white matter, small vessels, and large vessels were generated. Tissue-specific analysis showed a signal change rate of 4.53% in gray matter, 2.98% in white matter, 5.79% in small vessels, and 7.24% in large vessels. Different temporal patterns as well as different levels of activation were identified in the functional response from various types of tissue. High correlation exists between cortical gray matter and subcortical white matter (r = 0.957), while the vessel behaves somewhat different temporally. The cortical gray matter fits best to the assumed input function (r = 0.957) followed by subcortical white matter (r = 0.829) and vessels (r = 0.726). The automated algorithm of tissue-specific analysis thus can assist functional MRI studies with different modalities of response in different brain regions.     

Tissue segmentation-assisted analysis of fMRI for human motor response: an approach combining artificial neural network and fuzzy C means

The authors have developed an automated algorithm for segmentation of magnetic resonance images (MRI) of the human brain. They investigated the quantitative analysis of tissue-specific human motor response through an approach combining gradient echo functional MRI and automated segmentation analysis. Fifteen healthy volunteers, placed in a 1.5 T clinical MR imager, performed a self-paced finger opposition throughout the activation periods. T1-weighted images (WI), T2WI, and proton density WI were acquired for segmentation analysis. Single-slice axial T2* fast low-angle shot (FLASH) images were obtained during the functional study. Pixelwise cross-correlation analysis was performed to obtain an activation map. A cascaded algorithm, combining Kohonen feature maps and fuzzy C means, was applied for segmentation. After processing, masks for gray matter, white matter, small vessels, and large vessels were generated. Tissue-specific analysis showed a signal change rate of 4.53% in gray matter, 2.98% in white matter, 5.79% in small vessels, and 7.24% in large vessels. Different temporal patterns as well as different levels of activation were identified in the functional response from various types of tissue. High correlation exists between cortical gray matter and subcortical white matter (r = 0.957), while the vessel behaves somewhat different temporally. The cortical gray matter fits best to the assumed input function (r = 0.957) followed by subcortical white matter (r = 0.829) and vessels (r = 0.726). The automated algorithm of tissue-specific analysis thus can assist functional MRI studies with different modalities of response in different brain regions.     

Increased gray matter volume in lithium-treated bipolar disorder patients

Lithium's neurotrophic effects have been reported in several in vitro and ex vivo studies. Preliminary human studies with magnetic resonance imaging (MRI) and spectroscopy have recently provided evidence of lithium-induced increases in gray matter volumes and N-acetyl-aspartate levels. In order to further examine the hypothesis that lithium treatment would relate to detectable increases in gray matter brain content, we blindly measured gray and white matter volumes in MRI images of 12 untreated and 17 lithium-treated bipolar patients and 46 healthy controls. Using multivariate analysis of covariance with age and gender as covariates, we found that total gray matter volumes were significantly increased in lithium-treated (747.9 +/- 69.8 cm(3)) compared with untreated patients (639.2 +/- 91.2 cm(3); F = 10.6; d.f. = 1, 25; P = 0.003) and healthy individuals (675.8 +/- 61.8 cm(3); F = 17.4; d.f. = 1, 59; P < 0.001), suggesting in vivo effects of lithium on gray matter, which could possibly reflect lithium's neurotrophic effects.     

Differences in and correlations between cognitive abilities and brain volumes in healthy control,mild cognitive impairment,and Alzheimer disease groups

The purpose of this study is to investigate differences in and correlations between cognitive abilities and brain volumes in healthy control (HC), mild cognitive impairment (MCI), and Alzheimer's disease (AD) groups. The Korean Version of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD‐K), which is used to diagnose AD, was used to measure the cognitive abilities of the study subjects, and the volumes of typical brain components related to AD diagnosis—cerebrospinal fluid (CSF), gray matter (GM), and white matter (WM)—were acquired. Of the CERAD‐K subtests, the Boston Naming Test distinguished significantly among the HC, MCI, and AD groups. GM and WM volumes differed significantly among the three groups. There was a significant positive correlation between Boston Naming Test scores and GM and WM volumes. In conclusion, the Boston Naming Test and GM and WM brain volumes differentiated the three tested groups accurately, and there were strong correlations between Boston Naming Test scores and GM and WM volumes. These results will help to establish a test method that differentiates the three groups accurately and is economically feasible. Clin. Anat. 29:473–480, 2016. © 2016 Wiley Periodicals, Inc.     

基于中心化自动加权多任务学习的早期轻度认知障碍诊断

轻度认知障碍(MCI)是阿尔茨海默病(AD)的早期阶段,是治疗AD的最佳时期,因此对MCI的诊断非常重要。多模态数据可以全面分析疾病的状况,有利于疾病的准确诊断,但是现有方法并不能同时有效地分析多个模态数据之间的关系,无法有效结合功能态数据和结构态数据之间的优势。提出一种中心化自动加权多任务学习方法用于MCI的诊断。该方法可以同时学习不同模态的数据,有效地结合数据之间的优势。首先,分别对功能态数据rs-fMRI和结构态数据DTI构造脑网络;其次,基于多模态数据设计新的多任务特征学习模型,每个任务的重要性和模态之间的平衡关系会被自动学习,包括不同模态间的相似性和特异性,以获得稳定且有识别力的表达特征;最后,将选取的特征输入支持向量机模型进行分类诊断。实验基于Alzheimer′s Disease Neuroimaging Initiative(ADNI)公共数据库,包括明显记忆问题(SMC)、早期轻度认知障碍(EMCI)、晚期轻度认知障碍(LMCI)和正常受试者(NC)。所提出的方法对于NC vs SMC、SMC vs EMCI、SMC vs LMCI和EMCI vs LMCI等4种不同类型数据,诊断结果分别为76.67%、79.07%、80.56%和74.29%,与其他传统算法相比,分类准确率都有明显的提高,有望应用于对早期轻度认知障碍的诊断分析。     

White matter lesions are associated with cortical atrophy more than entorhinal and hippocampal atrophy

The goal of this study was to examine the relationship between subcortical vascular disease and brain atrophy in patients with Alzheimer's disease (AD) and mixed dementia (i.e., AD and subcortical vascular disease together). MRI was performed on 77 cognitively normal (CN) subjects, 50 AD and 13 mixed dementia patients. Subcortical vascular disease was determined by white matter hyperintensities (WMH) volume and presence of subcortical lacunes. Brain atrophy was measured using total brain cortical gray matter (CGM), entorhinal cortex (ERC) and hippocampal volumes. CGM volume, but not ERC or hippocampal volume was inversely related to WMH volume in patients and controls. In contrast, no relationship was detected between CGM, ERC, or hippocampal volumes and subcortical lacunes. Furthermore, no interaction was found between WMH and diagnosis on cortical atrophy, implying that WMH affect cortical atrophy indifferently of group. These results suggest that subcortical vascular disease, manifested as WMH, may affect cortical atrophy more than ERC and hippocampal atrophy. Further, AD pathology and subcortical vascular disease may independently affect cortical atrophy.     

基于极限学习机的阿尔兹海默病辅助诊断

阿尔兹海默病是一种渐进发展式的痴呆疾病, 其脑部随着病情发展逐渐出现萎缩。利用磁共振脑图像解剖学特征的变化, 提出一种使用极限学习机来诊断阿尔兹海默病以及轻度认知障碍的方法。采用FreeSurfer软件, 分析从ADNI数据库的818份磁共振图像中得到的脑部解剖学特征。首先对这些特征使用线性回归模型来估计正常衰老引起的萎缩因素, 并将其从特征中去除;随后采用极限学习机作为分类器, 使用处理后的特征来诊断阿尔兹海默病和轻度认知障碍。在实验过程中, 通过十折交叉验证来测试该方法的诊断准确率、敏感度、特异度和曲线下面积。通过100次实验求平均的方式计算得出, 该方法诊断阿尔兹海默病的准确率达到87.62%, 曲线下面积达到94.25%;诊断轻度认知障碍的准确率达到73.38%, 敏感度达到83.88%, 其中年龄矫正能有效提高轻度认知障碍诊断的准确率。实验结果表明, 该方法能有效诊断阿尔兹海默病和轻度认知障碍。     

强迫症患者大脑灰质体积变化——基于体素的形态学分析

目的探讨强迫症(OCD)患者大脑灰质体积的变化,并分析其在发病过程中可能存在的相关机制。方法选择31例年龄17～47岁重度强迫症患者和31例正常对照被试者,获取脑结构磁共振T1图像,使用基于体素的形态学测量(VBM)方法,比较强迫症组和对照组大脑灰质体积的差异,并将患者灰质体积差异区与其临床评分进行相关分析。结果与对照组相比,OCD患者在左侧壳核、岛叶、运动前区、顶上小叶以及右侧角回处体积显著减小。左侧壳核和岛叶的体积与患者贝克焦虑量表(BAI)评分成显著负相关。结论左侧壳核、岛叶、运动前区、顶上小叶以及右侧角回的灰质体积变化影响了该脑区功能,从而导致了OCD患者的部分症状。其中左侧壳核以及岛叶的损伤与患者焦虑情绪的异常密切相关。