Signal processing and Bio-medical Imaging Lab, Department of Electronics and Communication Engineering, Indraprastha Institute of Information Technology-Delhi (IIIT-D), India.
Department of Neuroradiology, Neurosciences Centre, All India Institute of Medical Sciences (AIIMS), Delhi, India.
Med Image Anal. 2017 Dec;42:228-240. doi: 10.1016/j.media.2017.08.007. Epub 2017 Aug 30.
Motivated by recent interest in identification of functional brain networks, we develop a new multivariate approach for functional brain network identification and name it as Multivariate Vector Regression-based Connectivity (MVRC). The proposed MVRC method regresses time series of all regions to those of other regions simultaneously and estimates pairwise association between two regions with consideration of influence of other regions and builds the adjacency matrix. Next, modularity method is applied on the adjacency matrix to detect communities or functional brain networks. We compare the proposed MVRC method with existing methods ranging from simple Pearson correlation to advanced Multivariate Adaptive Sparse Representation (ASR) methods. Experimental results on simulated and real fMRI dataset demonstrate that MVRC is able to extract functional brain networks that are consistent with the literature. Also, the proposed MVRC method is 650-750 times faster compared to the existing ASR method on 90 node network.
受近期对识别功能大脑网络的兴趣的推动，我们开发了一种新的多变量方法来识别功能大脑网络，并将其命名为基于多变量向量回归的连接性 (MVRC)。所提出的 MVRC 方法同时回归所有区域的时间序列到其他区域的时间序列，并考虑其他区域的影响来估计两个区域之间的成对关联，并构建邻接矩阵。接下来，模块性方法应用于邻接矩阵上，以检测社区或功能大脑网络。我们将所提出的 MVRC 方法与现有的方法进行了比较，范围从简单的 Pearson 相关到先进的多变量自适应稀疏表示 (ASR) 方法。在模拟和真实 fMRI 数据集上的实验结果表明，MVRC 能够提取与文献一致的功能大脑网络。此外，与现有的 ASR 方法相比，在 90 个节点网络上，所提出的 MVRC 方法的速度快 650-750 倍。
