CASIA OpenIR  > 毕业生  > 博士学位论文
有监督的多模态数据融合方法及其在脑疾病影像学标记检测中的应用
戚世乐
学位类型工学博士
导师隋婧
2017-11
学位授予单位中国科学院研究生院
学位授予地点北京
关键词有监督的多模态融合 精神分裂症 抑郁症 认知损伤 Microrna132
摘要多模态脑影像能够从不同的角度反映大脑的功能或者结构信息。尽管从扫描
对象采集多种模态的影像数据已经比较普遍,实际操作中却常常被分开单独分析,
或者是对各个磁共振影像
(MRI)数据的分析结果进行简单比较或相关,模态间的
互信息
(cross information)无法得到充分利用。最近的研究表明,结合多种模态的
影像数据不仅能够从多个角度理解人脑,并且具有一定的互补性,能够发现与脑
疾病有关的多种模态之间的共变性
, 这是单模态分析无法实现的。目前已经有较
多功能的
(fMRI)或者结构的(sMRI, dMRI)影像学研究尝试以无监督的数据驱动
方式挖掘神经影像学标记物,以加深对于脑疾病患者的认知障碍和相关神经病理
机制的理解。除此以外, 充分利用多模态影像数据、临床和行为学数据,挖掘与
感兴趣指标密切相关的脑影像共变模式,并以此来实现认知能力与治疗疗效的个
体化预测,则是当前研究的热点和难点。
本论文致力于开发一种有监督的、多变量脑影像学融合方法,能够充分利用
临床感兴趣的指标如认知、症状、行为、基因变量等作为先验参考信息引导多模
MRI 影像的联合分析,从而精准地挖掘出与特定临床指标显著关联的多模态
MRI 共变模式,并以此开展进一步的影像学标记对认知、行为等的预测。本文的
主要创新性工作包括:

(1)提出一种有监督的多模态融合方法 MCCAR+jICA (multi-site canonical
correlation analysis with reference + joint independent component analysis)
,该方法
能够基于参考信息挖掘与之显著相关的多模态共变模式。 在模拟实验中,我们从
多个层面对比了
MCCAR+jICA 及它的多种变形方法,结果表明 MCCAR+jICA
能够更准确的提取出与先验参考信息相关的影像成分,并且达到更高的估计精度。
在真实数据实验中, 我们以工作记忆评分作为参考信息来探索精神分裂症中与工
作记忆损伤显著相关的
3 模态共变模式。 结果表明分数低频振荡振幅(fALFF)
的中央执行网络、 灰质体积
GM 中的突显网络、 分数各项异性 FA 中的胼胝体白
质纤维束等作为三模态共变的
(fMRI-sMRI-dMRI)脑网络模式不仅与工作记忆评
分密切相关,而且在健康对照和病人之间存在显著差异,更重要的是该三模态共
变模式能够在独立数据集中高度复现,验证了
MCCAR+jICA 方法的有效性和广
泛的应用前景。

(2)研究与认知功能损伤密切相关的磁共振影像学标记,实现个体化多子域
认知水平的预测。认知功能损伤被认为是包括精神分裂症在内的多种脑疾病广泛

存在的功能障碍, 基于提出的有监督融合方法 MCCAR+jICA、多种认知评分(
分和注意力、工作记忆、语言学习三项认知子域评分
),和两个独立的三模态 MRI
数据集, 本工作对精神分裂症与认知损伤相关的多模态共变脑网络进行了深入
挖掘和多重验证。 结果表明:
a)与精神分裂症的认知损伤相关的影像标记网络
(neuromarker network)包括突显网络、执行控制网络和后默认网络, 该核心三网
络具有模态特异性和共変性
, 并且在独立数据集可复现; b) 通过比较 4 个认知
子域对应的影像学标记网络,发现结构影像特征
(GM/FA)的空间分布在子域间高
度一致,而功能影像特征
(fALFF)则在不同子域存在较大的分布差异,即对不同
认知功能
(任务)间的细微差异反映更敏感; c) 更重要的是,检测得到的核心多模
态特征能够较好的预测个体化认知总分
(r=0.463), 并且能够泛化到多种认知功能
(子域)的预测。
(3)首次以 microRNA 为参考信息,探索抑郁症中与 miR-132 异常表达相关
的多模态脑影像
(fMRI-sMRI-dMRI)共变模式,这是一种影像基因关联研究的新
方式。通过将血液中
miR-132 的表达量(抑郁症>正常人)作为参考信息来引导 3
MRI 特征的融合,我们发现: a) 抑郁症中高表达的 miR-132 与前额叶边缘
系统降低的
fALFF GM 相关; b)高表达 miR-132 相关的脑区与抑郁症的注意
力和执行功能损伤密切相关。上述关联关系在未用药
MDD 全体和从未用药的子
(drug-naïve MDDs)中均一致存在。这些发现加深了我们对 MDD miR-132
常表达与前额叶
边缘系统以及认知损伤的理解,而它们之间内在的因果关系有
待于进一步验证。

其他摘要Noninvasive neuroimaging has provided remarkable new insights into human
brain structure and function in both health and disease. There is increasing evidence
that instead of using a single imaging modality to study its relationship with
physiological or cognitive features, people are paying more attention to multimodal
fusion, an approach that is able to capitalize on the strength of multiple imaging
techniques, since it can uncover the hidden relationships that might be missed from
separate unimodal imaging studies. Compelling evidence has confirmed that
neuropsychiatric disorders reflect fundamental differences in brain structure and
function. By jointly analyzing rich types of data and taking advantage of the crossinformation, multimodal fusion can help better reveal the potential functional-structural
covariations. For example, how brain structure shapes brain function, and to what
degree brain function is related to the underlying brain anatomy, can be examined.
Increasingly, studies are focusing on identifying the intrinsic functional or structural
brain patterns that may ultimately drive a specific domain of human cognition or
behavior, whereas most existing fusion models are purely data-driven. Hence in this
work, we are motivated to develop a supervised multimodal fusion method that is able
to discover the co-varying imaging patterns particularly related to a referred
measurement more precisely and robustly.
The main achievements of our study are as follows:
(1) We proposed a fusion with reference model called “multi-site canonical
correlation analysis with reference + joint independent component analysis”
(MCCAR+jICA), which can precisely identify co-varying multimodal imaging patterns
closely related to the reference, such as cognitive scores. In a three-way fusion
simulation, the proposed method was compared with its alternatives on multiple facets;
MCCAR+jICA outperforms others with higher estimation precision and high accuracy
on identifying a target component with the right correspondence. In human imaging
data, working memory performance was utilized as a reference to investigate the co-

varying working memory associated brain patterns among three modalities and how
they are impaired in schizophrenia. Two independent cohorts were used. Similar brain
maps were identified between the two cohorts along with substantial overlaps in the
central executive network in fMRI, salience network in sMRI, and major white matter
tracts in dMRI. These regions have been linked with working memory deficits in
schizophrenia in multiple reports and MCCAR+jICA further verified them in a
repeatable, joint manner, demonstrating the ability of the proposed method to identify
potential neuromarkers for mental disorders.
(2) Cognitive dysfunction has been recognized as a primary and enduring core
deficit in many psychotic disorders, including schizophrenia. Imaging neuromarkers
that interrelate cognitive impairments and can ultimately predict individual cognitive
performance are a major focus in the new era of psychiatric study. To this end, we utilize
our supervised fusion approach to identify multimodal brain networks that are closely
associated with cognitive impairment. We combine three types of magnetic resonance
imaging features guided by cognitive scores in a discovery dataset and replicated in an
independent dataset. Domain scores of attention, working memory and verbal learning
were used as references to identify common and domain-specific multimodal
neuromarkers in health individuals and schizophrenia patients. Results show structure
features are more consistent cross different cognitive domains, while functional features
are more sensitive to cognitive domain differences. Three multimodal networks: the
salience network, the central executive network and the posterior default mode network
may play a crucial role as anatomical substrates of neurocognition. These networks
achieved high accuracy for predicting the cognitive scores (r=0.463) which can be
generalized to predict multiple cognitive performance. This work defines modalityspecific brain networks that may be broadly applicable as neuromarkers of cognitive
impairment.

(3) This work will be the first attempt to investigate how microRNA132
dysregulation may impact covariation of multimodal brain imaging data in 81
unmedicated major depressive patients and 123 demographically-matched healthy
controls, as well as in a medication-naïve subset of major depressive patients.
MicroRNA132 values in blood (patients>controls) was used as a prior reference to

guide fusion of three MRI features, i.e., fractional amplitude of low frequency
fluctuations, gray matter volume, and fractional anisotropy. The multimodal
components correlated with microRNA132 also show significant group difference in
loadings. Results indicate: 1) Higher microRNA132 levels in major depressive disorder
are associated with both lower fractional amplitude of low frequency fluctuations and
lower gray matter volume in fronto-limbic network;
2
) The identified brain regions
linked with increased microRNA132 levels were also associated with poorer cognitive
performance in attention and executive function.
All of the above associations are
consistent on either unmedicated or medication-naïve patients. These findings support
that the frontal-limbic system previously implicated in major depression in both
functional, anatomical and structural brain-imaging studies, were associated with miR-
132 dysregulation, while the intrinsic causality between miR-132 and brain alterations
awaits further investigation and verification.

文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/20404
专题毕业生_博士学位论文
作者单位中国科学院自动化研究所
推荐引用方式
GB/T 7714
戚世乐. 有监督的多模态数据融合方法及其在脑疾病影像学标记检测中的应用[D]. 北京. 中国科学院研究生院,2017.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
有监督的多模态数据融合方法及其在脑疾病影(8575KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[戚世乐]的文章
百度学术
百度学术中相似的文章
[戚世乐]的文章
必应学术
必应学术中相似的文章
[戚世乐]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。