Using multivariate pattern analysis (MVPA) to determine learning related changes in structural brain connectivity with diffusion MRI

记忆依赖于全脑记忆系统网络的相互作用。在学习过程中，突触连接的重塑被启动。在随后的巩固阶段，进一步的突触和系统调制使新的记忆稳定并将其整合到现有的网络中。扩散加权磁共振成像方法的最新发展表明，在活体内可以观察到由学习引起的快速结构变化。最初的研究和我们自己的初步实验表明，在学习60-90分钟后，就可以在微观结构(灰质)和宏观结构(白质)水平上检测到大脑连通性的变化。因此，研究依赖学习的变化为研究大脑结构如何引起神经功能和认知提供了一个独特的机会。由于连通性数据的高度多变量性质，我们将采用机器学习方法(多变量模式分析，MVPA)进行统计假设检验。我们已经在高密度的通宵睡眠脑电数据中成功地实现了该方法，并将在实验设计、信号预处理和特征选择方面对扩散的MRI数据进行改进。我们的目标是研究在学习后的几个小时和几天内，系统记忆巩固后大脑连接的局部和网络变化。特别是，我们将观察睡眠和清醒时大脑连通性的发展。相反，我们还将调查个体的连接体如何预测成功的学习。总而言之，这个项目将使我们能够确定大脑结构与学习和记忆的神经生物学相关性，同时推动MVPA在高度多变量数据中进行假设检验的使用。

Memory depends on the interaction of brain-wide networks of memory systems. During learning, remodeling of synaptic connectivity is initiated. During following consolidation periods, further synaptic and systems modulations render the new memory stable and integrate it into existing networks. Recent developments in diffusion-weighted magnetic resonance imaging methods give indication that rapid structural changes induced by learning can be observed in humans in vivo. First studies and our own preliminary experiments have shown that changes in brain connectivity can be detected on a microstructural (grey matter) and macrostructural (white matter) level already after 60-90 min of learning. Studying learning-dependent changes thus provide a unique opportunity to investigate how brain structure gives rise to neural function and cognition. Because of the highly multivariate nature of connectivity data, we will adopt a machine learning approach (multivariate pattern analysis, MVPA) for statistical hypothesis testing. We already successfully implemented this method in high-density all-night sleep EEG data, and we will adapt experimental design, signal preprocessing, and feature selection to diffusion MRI data. We aim at investigating the local and network changes in brain connectivity following systems memory consolidation during the hours and days after learning. Particularly, we will look at the development of brain connectivity during sleep and wakefulness. Conversely, we will also investigate how the individual’s connectome predicts successful learning. Together, this project will allow us to determine the neurobiological relevance of brain structure for learning and memory, and at the same time advance the use of MVPA for hypothesis testing in highly multivariate data.