Structural Brain Network Characteristics Can Differentiate CIS from Early RRMS
Structural Brain Network Characteristics Can Differentiate CIS from Early RRMS
复制标题
DOI：
10.3389/fnins.7016.00014
复制
发表时间：
2016-02-02
影响因子：
4.3
通讯作者：
Groppa, Sergiu
中科院分区：
文献类型：
--
作者：
Muthuraman, Muthuraman;Fleischer, Vinzenz;Groppa, Sergiu
局灶性脱髓鞘病变、弥漫性白质(WM)损害和灰质(GM)萎缩直接影响多发性硬化患者的疾病进展。本研究的目的是与早期复发-缓解型多发性硬化症(RRMS)患者比较，确定临床隔离综合征(CIS)患者GM和WM结构网络的特异性特征。对20例CIS患者、33例RRMS患者和40例健康人进行3T-MRI检查。应用扩散张量成像，结合WM的概率束成像和分数各向异性(FA)图，以及GM的皮质厚度相关分析，确定结构连通性模式。借助于图论方法进行网络拓扑分析，以在不同的社区级别(模块化、聚类系数、全局和局部效率)表征网络。最后，应用支持向量机(Support Ventures Machine，简称支持向量机)自动区分两组样本。与CIS受试者相比，RRMS患者被发现具有更高的模块连接性和更高的局部聚集性，突出了GM和WM的局部处理增加。与健康对照组相比，两组都表现出更高的模块化和聚集系数。支持向量机算法使用GM得到的聚类系数作为分类器，获得了97%的准确率，使用概率纤维束成像得到的WM获得了65%的准确率，根据FA图的模块化程度获得了67%的准确率来区分CIS和RRMS患者。我们证明，与CIS和健康受试者相比，早期RRMS患者的模块化和局部连接性明显增加。仅基于一次解剖扫描，在没有先验信息的情况下，我们开发了一种自动化的、独立于研究人员的范例，可以准确地区分具有这些临床相似疾病实体的患者，从而可以补充目前临床诊断的及时传播标准。
Focal demyelinated lesions, diffuse white matter (WM) damage, and gray matter (GM) atrophy influence directly the disease progression in patients with multiple sclerosis. The aim of this study was to identify specific characteristics of GM and WM structural networks in subjects with clinically isolated syndrome (CIS) in comparison to patients with early relapsing-remitting multiple sclerosis (RRMS). Twenty patients with CIS, 33 with RRMS, and 40 healthy subjects were investigated using 3 T-MRI. Diffusion tensor imaging was applied, together with probabilistic tractography and fractional anisotropy (FA) maps for WM and cortical thickness correlation analysis for GM, to determine the structural connectivity patterns. A network topology analysis with the aid of graph theoretical approaches was used to characterize the network at different community levels (modularity, clustering coefficient, global, and local efficiencies). Finally, we applied support vector machines (SVM) to automatically discriminate the two groups. In comparison to CIS subjects, patients with RRMS were found to have increased modular connectivity and higher local clustering, highlighting increased local processing in both GM and WM. Both groups presented increased modularity and clustering coefficients in comparison to healthy controls. SVM algorithms achieved 97% accuracy using the clustering coefficient as classifier derived from GM and 65% using WM from probabilistic tractography and 67% from modularity of FA maps to differentiate between CIS and RRMS patients. We demonstrate a clear increase of modular and local connectivity in patients with early RRMS in comparison to CIS and healthy subjects. Based only on a single anatomic scan and without a priori information, we developed an automated and investigator-independent paradigm that can accurately discriminate between patients with these clinically similar disease entities, and could thus complement the current dissemination-in-time criteria for clinical diagnosis.