工作记忆(working memory, WM)是人类几乎所有高级认知功能的基石，涉及到大脑众多脑区的神经活动。本研究基于大被试量的功能磁共振数据，探究了工作记忆的认知神经基础，从脑功能区激活与脑网络连接两方面展开研究。 首先，本研究刻画了工作记忆的脑功能图谱。尽管脑功能区的个体间变异普遍存在，但少有研究关注人类大脑中工作记忆脑功能区的个体差异。本研究第一部分的主要目的是基于个体特异的脑功能区，量化工作记忆功能激活区的个体差异，构建工作记忆的脑功能图谱。研究采集了477名健康成人被试完成工作记忆任务时的fMRI数据，随后两名评分者采用半自动化方法独立标定了每名被试特异的工作记忆脑功能区，包括了双侧的额极(frontal pole, FP)、额中回(middle frontal gyrus, MFG)、额叶眼动区(frontal eye field, FEF)、顶上小叶(superior parietal lobule, SPL)、脑岛(insular, INS)、楔前叶(precuneus, PCUN)和前扣带回(anterior cingulate cortex, ACC)脑区。基于此，本研究构建了工作记忆的脑功能图谱，描述了工作记忆功能区重要的个体间变异，并且分析了脑区的半球偏侧化和性别差异。同时，本研究定义的工作记忆脑功能图谱与元分析图谱进行了比较，证明了功能图谱的可靠性；还分析了图谱定义的脑区激活强度与工作记忆行为表现的相关，证实了图谱的可用性。综上，这一部分研究基于个体差异构建了工作记忆的脑功能图谱，为标准化功能区定位和数据整合提供了重要的功能和空间参考。 进一步，本研究分析了工作记忆网络的功能连接。人类完成工作记忆任务时，大脑多个脑区同时激活并构成了功能网络。然而，目前尚不清楚任务状态下工作记忆网络的特异性改变。本研究第二部分结合了静息和任务态fMRI数据，探究了工作记忆网络在任务状态与静息状态功能连接的差异，从网络层面进一步分析了工作记忆的神经基础。研究采集了169名健康被试的静息和N-back任务数据，并进行了功能连接分析。结果发现，相比静息状态，工作记忆网络在任务状态下连接强度(connectivity strength)增加，而模块性(modularity)下降，证明了任务驱动的工作记忆网络整合程度的变化。进而，研究通过聚类将工作记忆网络连接拆分为模块内连接(intra-module connectivity)与模块间连接(inter-module connectivity)，发现了任务态下模块内连接的减弱和模块间连接的增强。并且，模块内连接还与N-back行为表现显著关联，进一步阐明了工作记忆网络的变化及行为学意义。因此，第二部分研究基于静息和任务态工作记忆网络连接的比较，发现了模块内和模块间连接的不同变化，便于理解人类工作记忆网络的连接特性。 通过上述研究，我们揭示了工作记忆的脑功能区的个体差异，构建了脑功能区图谱，并分析工作记忆网络的功能连接特性，为理解工作记忆的神经基础提供了新的见解。

Working memory (WM) system involves in almost all complex high-level cognitive abilities and contains the neural activation of an array of regions in human cortex. Here we investigated the neural basis of WM with a large cohort of participants from the points of region’s activation and network’s connectivity in human cortex. First, we constructed functional probabilistic atlas of WM. As little is known about dramatic individual differences in WM activated regions, the goal of the present study was to quantitate individual variability in WM regions, and construct functional probabilistic atlas. To do so, functional magnetic resonance imaging (fMRI) data of 477 healthy volunteers were obtained and two raters independently marked WM regions (including the frontal pole (FP), middle frontal gyrus (MFG), frontal eye field (FEF), superior parietal lobule (SPL), insular (INS), precuneus (PCUN) and anterior cingulate cortex (ACC)) for each participant using a semi-automated method. Functional probabilistic atlas of WM was delineated to characterize individual variability, revealing that the WM regions were highly variable across participants. Inter-hemisphere asymmetry and gender differences were also found in most regions. Moreover, our results indicated that probabilistic map was highly consistent with WM meta-analytic atlas, showing the reliability of probabilistic map. And we also evaluated usability of probabilistic atlas by measuring associations between activation magnitudes in WM regions and N-back behavioral performances. Overall, present WM functional probabilistic atlas can provide a robustly functional and spatial reference for standardizing functional localization and data integration. Next, we investigated modulation of task-related functional connectivity in WM network. Multiple regions simultaneously activated during WM process and organized into a functionally specific WM network. However, it is unknown whether there is task-specific difference in WM network connectivity comparing with intrinsic connectivity. To do so, we combined resting and task-state fMRI from 169 healthy participants and calculated WM network functional connectivity (FC). We found that WM network modularity (a measure of network integration) reduced during n-back task, while network FC strength increased. This outcome alluded to greater interaction within WM network during WM task. Then we examined intra-module and inter-module connectivity in WM network during task and rest states. We found reversed patterns, that is, task-state inter-module FC strength increased, whereas intra-module FC strength during task condition was reduced. Further, intra-module connectivity was associated with WM ability. Thus, our results shed light on the differences of WM network during task and rest state, and highlighted dissociation between intra- and inter-module FC in supporting task cognitive processing. In sum up, we demonstrated individual difference in WM activation regions and constructed functional probabilistic atlas. We further analyzed WM network’s connectivity in different states, and provided more insight into the neural basis underlying WM process.