基于磁共振影像的有监督多变量生物标志鉴定方法

Identifying neuroimaging markers of brain diseases based on data mining techniques has been a hot topic today. In this project, we will develop advanced multivariate data fusion model based on supervised learning. The proposed method will take full advantage of the richness of multimodal magnetic resonance imaging data as well as the clinical information, including cognitive/ behavioral/symptomatic scores and genetic data. Specifically, we will develop a supervised, goal-directed model that incorporates prior clinical information as a reference to guide the multimodal MRI data fusion, in order to discover the neuroimaging markers that are particularly associated with the specific clinical/cognitive measurement. These neuroimaging markers may be used further for disease classification and individualized prediction. We will also released the relevant tools on supervised fusion. These methods and tools will be applied to two practical clinical cases：1) Identify the multimodal MRI imaging markers that are specifically associated with the sustained attentional deficits for ADHD children and their healthy comparisons. 2) Uncover the brain imaging mapping closely related to the abnormal microRNA 132 levels in major depressive disorders (MDD). This may be promising avenues to identify objective biomarkers, improve understanding of the pathophysiology of mental disorders, and thus lead to more appropriate clinical care. Accomplishment of this project will provide valuable tools and methods for the community in exploring goal-directed neuroimaging markers based on multimodal MRI data, which may have been missed by existing multivariate approaches, suggesting significant translational impact on "precision medicine". In summary, this work aims at the international scientific frontiers, and is of profound medical benefits.

基于数据挖掘开展脑疾病影像学标志鉴定是当今研究的热点和前沿。本课题将开发有监督的、多变量影像标志鉴定方法，充分利用多模态磁共振影像数据的丰富性，将被试的认知、行为、基因等临床指标作为先验参考信息，辅助多模态脑影像的联合源分离，实现有针对性的、精准地挖掘出与某一特定兴趣指标显著相关的多模态共变模式，从而加深对复杂脑疾病生理病理机制的认识。上述方法还将用于两项实际的临床分析: 1)考察与ADHD儿童持续注意力缺陷显著相关的多模态MRI影像学指标；2)挖掘与重度抑郁患者的microRNA132的异常表达水平密切相关的多种MRI特征组合；为进一步开展相关疾病分类、临床指标的个体化预测奠定基础。该项目的顺利完成将为界内提供一系列新方法，发布相关的工具箱，方便广大医生和科研工作者根据自己的特殊需求寻找感兴趣的影像学标志；为“精准医疗”这一重大研究热点做出贡献，具备广泛的应用前景和重大的转化医学价值

本项目开发了一系列有监督的、多变量影像标志挖掘方法，将被试的认知、行为、基因等临床指标作为先验参考信息，提出了基于MCCAR+jICA的多模态脑影像的联合源分离算法，能够有针对性地挖掘出与某一特定兴趣指标显著相关的多模态稳定共变模式。项目还提出了加入时间信息的平行ICA方法，能够将3D结构像与4D功能磁共振影像等不同维度的脑影像特征进行融合分析。此外，基于磁共振影像学特征，我们还提出了多种神经影像预测分类技术与方法，包括多尺度图卷积网络(能够整合从粗到细的脑连接信息)和针对fMRI时间序列的多尺度递归神经网络(能够整合体素级别的脑功能时间动态变化)，使得发现的影像学标志可以服务于临床的脑疾病风险或疗效预测，并应邀在权威期刊Biological Psychiatry撰写了综述。上述方法被应用于三项主要的临床分析: 1)考察与ADHD儿童的抑制控制能力失调显著相关的多模态MRI影像学指标；2)挖掘与重度抑郁患者的ECT治疗疗效反应密切相关的多种MRI特征组合，并进一步开展精细的亚型研究；3) 成功地基于MRI 预测个体认知能力和青少年罹患多种精神疾病的风险。目前本项目已经在Nature Communications，Biological Psychiatry等期刊发表SCI论文45篇，国际会议论文7篇，申请专利3项，获得授权2项，为界内提供一系列新的融合分析与预测方法理论，并集成发布于FIT工具箱, 具备重要的应用前景和重大的转化医学价值。

内容获取失败，请点击重试