毕业生知识库

脑图谱是脑科学研究的重要工具，精准可靠的脑图谱能极大地促进脑科学的发展。脑科学研究逐步深入的过程，也是脑图谱绘制从粗糙走向精细的过程。脑网络组图谱（Brainnetome Atlas），作为具有代表性的人类脑图谱之一，以连接组学的概念从宏观尺度绘制了全脑 246 个精细亚区。每个亚区具有独特的功能和结构连接模式，为研究脑结构和功能组织模式提供了重要的工具。当前的脑网络组图谱主要是基于群体脑影像的分析获得的，这使得它在实际应用中仍面临巨大的挑战。许多研究表明人脑功能和结构存在显著的个体差异性，而且这些差异无法在群体脑影像信息中得到体现。这使得群体水平的脑网络组图谱无法精确地应用在个体水平的脑疾病研究中，阻碍了基于脑图谱的个体化、精准化医疗的发展。针对这一现状，本文提出了稳定鲁棒的个体化脑图谱绘制方法，并将个体化脑图谱应用在疾病诊断和治疗规划中，有望促进个体化脑网络组图谱的研究及其在脑疾病诊疗中应用。本文的主要工作内容和创新点归纳如下：

脑图谱个体化绘制方法研究：通过分析结构和弥散磁共振影像，从解剖连接的角度开发了基于群组先验的脑图谱个性化的绘制方法（BAI-Net）。基于脑网络组图谱的先验信息，BAI-Net方法将纤维束连接指纹引入到个体化方法中，并利用图卷积模型对连接的空间相似性进行约束从而得到个体特异的脑分区结果。BAI-Net得到的个体脑分区可以在多次采集的影像上保持高可重复性，而且在多种磁共振扫描仪的影像上保持较高鲁棒性。相比之前的个体化方法，通过 BAI-Net 得到的脑区拓扑结构有更显著的个体间拓扑差异性。脑区个体间拓扑差异的空间分布符合已知人脑功能梯度的分布规律。BAI-Net个体脑区拓扑结构和超过30种个体认知指标显著有关。而且个体脑区拓扑分布具有一定程度的遗传性，即在陌生人之间和同卵双生子之间具有显著的拓扑差别。个体化脑网络组图谱为基于脑影像的连接组学分析提供了个体水平上的精确脑分区，也为精准治疗提供了有效的靶区定位。

基于个体化脑图谱的重度抑郁症诊断方法研究：重度抑郁症（MDD）广泛发生在人群中，但是以往基于群体脑图谱得到的全脑功能连接进行诊断结果并不理想，其在大样本上的分类准确较低。本研究利用BAI-Net方法重新定义的个体脑分区，并得到了个体特异的功能连接网络。相对于群体平均脑图谱得到的功能网络，基于个体特异的脑功能网络对重度抑郁症的诊断提升了4%的准确率，对汉密尔顿评分的预测相关性提升了101%。此外，我们发现脑分区本身的拓扑结构可以刻画重度抑郁症和正常人之间的差异，利用该拓扑异常特征的模型诊断准确率达70%。从脑区位置上显示，前额叶区域的异常拓扑分布对于重度抑郁症的预测能力最高。这种异常的脑分区拓扑结构可能提供了辅助的神经影像标志物，有望帮助重度抑郁症的精确诊断。

个体化脑图谱引导的精准经颅磁刺激方法研究：经颅磁刺激（TMS）是一种非侵入式的脑部刺激技术，已经广泛应用在脑疾病的刺激治疗中。个体化脑网络组图谱提供了个体特异的脑功能分区，可以用来引导经颅磁刺激的靶区位置。然而由于不同电导率的脑组织对于颅内磁感电场分布有巨大影响，精准经颅磁刺激方法需要定量评估不同线圈位姿下的颅内电场分布来优化对个体靶区的刺激效果。而以往基于计算电场仿真的优化方法并不能进行快速、准确地优化TMS线圈位姿。为了提升规划的及时性，本研究开发了快速个体线圈位姿优化方法（AnaRES-CPO）。相比于传统计算仿真优化方法需要5个小时以上处理时间，AnaRES-CPO方法可以在3分钟左右完成单个脑区位姿优化过程。通过在体运动区的经颅磁刺激，本研究验证了该方法在缩短时间的同时，与计算仿真模型的最优线圈位姿达到了相同水平的刺激效果。该方法能为脑网络组图谱引导的精准经颅磁刺激疗法提供高效的位姿规划。
 

Brain atlas is an important tool in brain science. Accurate and reliable brain atlas can greatly promote the development of brain science. With the development of  brain science,  the human brain parcellation are gradually  from rough to precise. Brainnetome atlas, as one of the representative human brain atlases, has parcellated 246 brain subregions based on the macro-scale connectome. Each subregion has a unique functional and structural connection pattern, which provide comprehensive understanding of the brain function and structure. However, it is still a great challenge to apply the Brainnetome atlas derived from the population images into individual brains. Many studies have shown there are significant individual variations on  human brain function and structure architecture, and these variations are not shown in the group brain images. Thus, it is difficult to apply the population-level Brainnetome atlas to the study of brain diseases and disorders at the individual level, hindering the research and development of atlas-based precision medicine. Based on this situation, this study developed a stable and robust individualized cerebral cartography method and applied the individualized brain atlas into the diagnosis of brain disorders and the stimulation planning, to hopefully improve the effectiveness and practicability of Brainnetome atlas in clinical diagnosis and treatment. The main contents and innovations of this paper are summarized as follows:

Individualized delineation method for Brainnetome atlas: Based on structure and diffusion MRIs, this study developed the brain atlas individualization network (BAI-Net) from the perspective of anatomical connectivity. Based on the group priors of the Brainnetome atlas, BAI-Net introduces the fiber-tract connectivity fingerprint into the individual brain cartographies, and uses the graph convolutional network to constrain the spatial similarity of the areal connectivity. The individual parcellations from the method can maintain high reproducibility on multiple scanned images from the same subject, and high robustness across different MRI scanners. Compared with the previous individualization method, the BAI-Net method has more significant topological differences among individuals. The spatial distribution of inter-subject areal topological differences consistent with the distribution of the functional gradient of human brain. Moreover, the BAI-Net topological structures are related with more than 30 individual cognitive and behavior scores. Moreover, this topological structure presents high heritability, indicating the significant topological differences between strangers and monozygotic twins. Individualized Brainnetome altas provides the individual-level brain cartography for image-based connectome analysis, and hopefully provides effective target location for precise treatment.

Diagnosis for major depression disorder based on individualized Brainnetome atlas: Major depression disorder(MDD) occurs widely in modern groups, but previous diagnosis based on the whole-brain functional connectome derived from the population brain atlas are relatively low. The classification precision on large samples remains low. This study redefined the individual brain regions through the BAI-Net method.  Compared with population-average functional connectome, this study found that it can be improved by 4% for the MDD classification precision based on individual-specific functional connectivity. And it also can be improved by 101% for the prediction correlation of Hanmilton score based on the individual-specific connectome. In addition, this study found that the topological structure of brain regions can also depict the differences between major depression disorder and healthy control. This classification precision is 70%. According to the spatial distribution of abnormal region, the abnormal topological distribution in the prefrontal region presents the greatest predictive power for major depression disorder. This abnormal topological structure of the brain region may provide an auxiliary neuroimaging marker to assist the precise diagnosis of major depression disorder.

Precise transcranial magnetic stimulation methods guided by individualized Brainnetome atlas: Transcranial magnetic stimulation (TMS) is a non-invasive brain stimulation technology, which has been widely used in regulating brain functional activities. Individualized brain atlas network method provides individual-specific brain parcellations, which can be used to guide the stimulation target in transcranial magnetic stimulation. However, due to the great influence of brain tissues with different conductivities on the distribution of magnetically induced electric field, it is necessary to  quantify  the distribution of intracranial electric fields under different coil positions to optimize the stimulation effect of the target area.  Previous  computation simulation models can not optimize the  TMS coil placement in a fast and precise way. In order to achieve the optimal stimulation for individual target, this study developed a fast individual coil placement optimization method (AnaRES-CPO) based on deep learning algorithm. Compared with the traditional computational simulation methods which takes more than 5 hours, this method can finish the optimization process for one brain target in around 3 minutes. The in-vivo transcranial magnetic stimulation showed that AnaRES-CPO method can accelerate the preprocessing time and achieve the same level of stimulation performance as the optimal coil placement derived from the computational simulation model. This method improves the computational efficiency for the precise TMS method guided by Brainnetome atlas.