Gansu Provincial Key Laboratory of Wearable Computing, School of Information Science and Engineering, Lanzhou University, Lanzhou, People's Republic of China.
Department of Neurology, Lanzhou University Second Hospital, Lanzhou, People's Republic of China.
J Neural Eng. 2020 Oct 23;17(5):056038. doi: 10.1088/1741-2552/abbc28.
It is important to improve identification accuracy for possible early intervention of major depressive disorder (MDD). Recently, effective connectivity (EC), defined as the directed influence of spatially distant brain regions on each other, has been used to find the dysfunctional organization of brain networks in MDD. However, little is known about the ability of whole-brain resting-state EC features in identification of MDD. Here, we employed EC by whole-brain analysis to perform MDD diagnosis.
In this study, we proposed a high-order EC network capturing high-level relationship among multiple brain regions to discriminate 57 patients with MDD from 60 normal controls (NC). In high-order EC networks and traditional low-order EC networks, we utilized the network properties and connection strength for classification. Meanwhile, the support vector machine (SVM) was employed for model training. Generalization of the results was supported by 10-fold cross-validation.
The classification results showed that the high-order EC network performed better than the low-order EC network in diagnosing MDD, and the integration of these two networks yielded the best classification precision with 95% accuracy, 98.83% sensitivity, and 91% specificity. Furthermore, we found that the abnormal connections of high-order EC in MDD patients involved multiple widely concerned functional subnets, particularly the default mode network and the cerebellar network.
The current study indicates whole-brain EC networks, measured by our high-order method, may be promising biomarkers for clinical diagnosis of MDD, and the complementary between high-order and low-order EC will better guide patients to get early interventions as well as treatments.
提高对重度抑郁症（MDD）早期干预的识别准确率非常重要。最近，有效连通性（EC），即空间上遥远的脑区之间的相互影响，已被用于寻找 MDD 中大脑网络功能失调的组织。然而，对于全脑静息态 EC 特征在 MDD 识别中的能力，我们知之甚少。在这里，我们采用全脑分析的 EC 来进行 MDD 诊断。
在这项研究中，我们提出了一种高阶 EC 网络，可以捕捉多个脑区之间的高级关系，以区分 57 名 MDD 患者和 60 名正常对照组（NC）。在高阶 EC 网络和传统的低阶 EC 网络中，我们利用网络特性和连接强度进行分类。同时，支持向量机（SVM）用于模型训练。通过 10 折交叉验证支持结果的推广。
分类结果表明，高阶 EC 网络在诊断 MDD 方面优于低阶 EC 网络，将这两种网络结合起来可以获得最佳的分类精度，准确率为 95%，敏感度为 98.83%，特异性为 91%。此外，我们发现 MDD 患者高阶 EC 的异常连接涉及多个广泛关注的功能子网，特别是默认模式网络和小脑网络。
本研究表明，通过我们的高阶方法测量的全脑 EC 网络可能是 MDD 临床诊断的有前途的生物标志物，高阶和低阶 EC 之间的互补性将更好地指导患者进行早期干预和治疗。
