MEG investigation of sleep and sleep-related memory reactivation

人们认为记忆在睡眠中通过重新激活而得到加强。动物实验表明，在学习过程中发现的神经元活动模式在睡眠中重演。在人类中，与学习相关的刺激可以提示重新激活，但自发重放很难确定。在这里，我们提出的实验中，我们记录MEG在长期记忆编码，工作记忆和睡眠，以检测模式的神经元活动，反映记忆的重新激活。首先，我们建议使用机器学习算法（特别是支持向量机，SVM）来检测和表征睡眠慢振荡和睡眠纺锤波过程中的自发记忆处理。接下来，我们将用深度卷积神经网络（CNN）来描述大脑活动的模式，这种模式可以区分工作记忆维持中的记忆内容。这种活动应该更类似于睡眠期间无刺激的再激活，而不是刺激锁定的编码活动。最后，我们将尝试使用这种活动模式来检测睡眠期间的记忆重放，使用工作记忆模式和自发睡眠活动之间的表征相似性。在独立分析中，我们还将研究与睡眠对记忆巩固的功能作用有关的睡眠生理学的某些方面。睡眠纺锤波（12 - 16 Hz）和慢振荡（< 1 Hz）以前被认为是全局过程，但最近的研究集中在它们的局部特性上。我们的初步数据表明，这两个过程都发生在局部，并传播到整个皮层。我们将使用源重建来描述这种本地生成和传播。我们建议使用脑磁图记录，因为它的优势相比，更传统的睡眠脑电图。MEG具有更好的信噪比，特别是在较高的频率下，并且大量的传感器（275个通道）允许从传感器空间到源空间的转换。在我们自己的初步数据和其他研究中，使用sLORETA的MEG数据源定位在整个皮质甚至皮质下区域被证明是可靠和准确的。因此，我们相信，我们将能够在睡眠期间定位记忆的重新激活，以及睡眠振荡的起源和传播，其精度比以前高得多。

Memories are supposed to be strengthened during sleep through reactivation. Animal experiments show that patterns of neuronal activity found during learning are replayed during sleep. In humans, reactivation can be cued with learning-related stimuli, but spontaneous replay has been difficult to pinpoint. Here, we propose experiments in which we record MEG during long-term memory encoding, working memory and sleep in order to detect patterns of neuronal activity that reflect memory reactivation. First, we propose to use machine learning algorithms (particularly support vector machines, SVM) to detect and characterize spontaneous memory processing during the sleep slow oscillation and the sleep spindle. Next, we will characterize the pattern of brain activity that distinguishes memory content in working memory maintenance with a deep convolutional neuronal network (CNN). This activity is supposed to be more similar to stimulus-free reactivation during sleep than stimulus-locked encoding activity. Finally, we will try to use this pattern of activity to detect memory replay during sleep using the representational similarity between the working memory pattern and spontaneous sleep activity. In independent analyses, we will also investigate some aspects of sleep physiology that are related to the functional role of sleep for memory consolidation. Sleep spindles (12 – 16 Hz) and slow oscillations (< 1 Hz) have previously been assumed to be global processes, but more recent studies focussed on their local properties. Our preliminary data indicate that both processes arise locally and propagate throughout the cortex. We will use source reconstruction to describe this local generation and propagation. We propose the use of MEG recordings, because of its advantages compared with the more conventional sleep EEG. MEG has a better signal-to-noise ratio especially in higher frequencies, and the high number of sensors (275 channels) allows transformation from sensor space to source space. Source localization of MEG data with sLORETA proved to be reliable and exact throughout cortical and even subcortical areas in our own preliminary data and other studies. We are therefore confident that we will be able to localize memory reactivation during sleep as well as origination and propagation of sleep oscillations with a much higher precision than has previously been possible.