Multi-variate and multi-modal modelling of neuroimaging data to better understand brain ageing

神经影像数据的多变量和多模式建模，以更好地了解大脑衰老

• 批准号: RGPIN-2020-05448
• 负责人: Chakravarty, Mallar
• 金额: $ 3.42万
• 依托单位: McGill University
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=740202
• 关键词: Multi; variate; multi; modal; modelling

Understanding variation in ageing requires sophisticated techniques that can examine brain structure and function across  volumetric, microstructural, and functional dimensions. Here, we leverage recent developments from my group for the integration of multi-modal MRI data (using non negative matrix factorization [NMF]) to build models that allow for an improved understanding of how variation in brain structure and function integrate and are related to cognition, with specific focus on the hippocampus. We will use publicly available data (the Human Connectome Project) containing state-of-the-art MRI of brain microstructure related myelin content and axonal properties in addition normative brain function using resting state functional MRI (rsfMRI). We will relate cognitive and functional tasks back to MRI-derived measures. Finally, we will use these techniques in a machine learning-based prediction model for individual age in a well-characterized healthy elderly sample. The goal would be to identify patterns of covariance that best predict cognitive function and age using an integrated analysis strategy that leverages NMF to define components of variance and then to examine the association of each component across cognitive measures using partial least squares (PLS). Compared to common decomposition techniques such as principal and independent component analysis, NMF outputs are more interpretable, sparse, and more spatially contiguous and non overlapping. Research Goals: 1) Microstructural parcellation of the human hippocampus and its relationship with cognition: NMF outputs individual weightings for each subject will be estimated to assess microstructural variability across the 330 unrelated study participants in HCP. Inputs to the NMF will be indices from structural MRI thought to reflect myelin content (T1/T2 ratio) and indices from diffusion MRI (fractional anisotropy and mean diffusivity). Subject-specific component weighting will be related back to 32 cognitive tests administered by the HCP using PLS. 2) Functional parcellation of the human hippocampus and its relationship with cognition: Here we will use a hippocampus-whole-brain approach to derive a rsfMRI based parcellation. Once again the relationship between our parcellation will be related back to the 32 cognitive measures using PLS. 3) Using NMF to predict individual age in a normative population: We will use existing data from the Whitehall II Imaging Sub-study, collected in the FMRIB Centre at Oxford. This dataset includes structural and diffusion MRI, as well as resting-state fMRI scans from 800 community-dwelling adults aged 60-85 years old. The development of an accelerated ageing biomarker has potential relevance to ageing studies in the future focused on heterogeneity of ageing trajectories and developing as a method that can help to better understand the variance in normative ageing. predicting future ageing trajectory.

了解衰老的变化需要复杂的技术来检查大脑的体积、微观结构和功能维度的结构和功能在这里，我们利用我的团队的最新进展来整合多模态 MRI 数据（使用非负矩阵分解 [NMF]）。 ）建立模型，以更好地理解大脑结构和功能的变化如何整合以及与认知相关，特别关注海马体。我们将使用包含以下内容的公开数据（人类连接组项目）。使用静息态功能性 MRI (rsfMRI) 来研究与髓鞘质含量和轴突特性相关的大脑微观结构以及正常的大脑功能。这些技术在基于机器学习的预测模型中针对特征明确的健康老年人样本中的个体年龄进行预测，目标是使用利用 NMF 来定义认知功能和年龄组成部分的集成分析策略来识别最能预测认知功能和年龄的协方差模式。方差和然后使用偏最小二乘法（PLS）检查认知测量中每个成分的关联，与主成分分析和独立成分分析等常见分解技术相比，NMF 输出更可解释、更稀疏，并且在空间上更连续且不重叠。 ：1）人类海马体的微观结构分区及其与认知的关系：NMF 输出每个受试者的个体权重将被估计，以评估 330 名不相关研究参与者的微观结构变异性HCP 的输入将是来自结构 MRI 的指数，该指数被认为反映了髓磷脂含量（T1/T2 比率），而来自扩散 MRI 的指数（分数各向异性和平均扩散率）将与进行的 32 项认知测试相关。 2) 人类海马体的功能分区及其与认知的关系：这里我们将使用海马体全脑方法来推导基于 rsfMRI 的分区之间的关系将再次与使用 PLS 的 32 种认知测量相关联。 3) 使用 NMF 来预测规范人群中的个体年龄：我们将使用来自 Whitehall II 成像子研究的现有数据，该数据集由牛津大学 FMRIB 中心收集，包括 800 名 60-85 岁社区居民的结构和扩散 MRI 以及静息态 fMRI 扫描。加速衰老生物标志物与未来的衰老研究具有潜在的相关性，重点关注衰老轨迹的异质性，并发展为一种有助于更好地理解预测未来衰老轨迹的规范差异的方法。