动态功能脑网络模型的多任务融合Lasso方法

发布时间：2018-02-03 13:32

  本文关键词： 静息态fMRI 动态功能脑网络 功能连接 多任务融合Lasso 稀疏 分类 阿尔兹海默症　出处：《中国图象图形学报》2017年07期 　论文类型：期刊论文

【摘要】：目的传统的静息态功能性磁共振成像(f MRI)的功能脑网络(FBN)研究是基于在整个扫描过程中FBN固定不变的假设。但是,最近的研究表明FBN是动态变化的,而且其中蕴含着丰富的信息。本文提出一种多任务融合最小绝对值收缩和选择算子(Lasso)方法来构建静息态f MRI的动态FBN。方法提出的多任务融合Lasso方法可以在构建动态FBN时,保留网络的稀疏性及子序列的时间平滑性。具体来说,首先用滑动窗方法得到交叠的静息态f MRI子序列;然后用多任务融合Lasso方法联合地估计一个样本的所有子序列的功能连接从而构建动态FBN,用k均值聚类算法得到每类样本子序列的功能连接的聚类中心,并将所有类的聚类中心组成回归矩阵;最后根据回归矩阵求样本的回归系数,将其作为特征进行分类,验证多任务融合Lasso方法对动态FBN建模的有效性。结果采用公开的f MRI数据集来验证多任务融合Lasso模型构建动态FBN的分类效果。实验使用阿尔兹海默症神经影像学计划(ADNI)公开的f MRI数据集中的阿尔兹海默症患者、早期轻度认知功能障碍患者和健康被试3组数据,并用准确率、灵敏度和特异度来评估算法的分类性能。在3组二分类实验中,本文方法分别达到了92.31%、80.00%和84.00%的准确率。实验结果表明,与静态FBN模型和其他传统的动态FBN模型相比,本文方法能取得更好的分类效果。结论本文提出的多任务融合Lasso构建动态FBN的方法,能有效地保留网络的稀疏性和子序列的时间平滑性,同时提高算法的分类效果,在一定程度上为脑部疾病的诊断提供帮助。多任务融合Lasso模型可以用于动态FBN的构建,挖掘功能连接的动态信息,同时整个算法可以用于基于f MRI数据的脑部疾病的分类研究中。
[Abstract]:Objective the traditional resting functional magnetic resonance imaging (fMRI) study is based on the assumption that FBN is fixed throughout the scanning process. Recent studies have shown that FBN is dynamic. And it contains abundant information. In this paper, we propose a multitask fusion minimum absolute contraction and selection operator Lasso). Methods to construct the dynamic MRI of resting f MRI. The multitask fusion Lasso method proposed by the method can be used to construct dynamic FBN. The sparsity of the network and the temporal smoothness of the subsequences are preserved. Firstly, the overlapping resting f MRI subsequences are obtained by the sliding window method. Then the multitask fusion Lasso method is used to estimate the functional connections of all subsequences of a sample to construct a dynamic FBN. By using k-means clustering algorithm, the cluster center of functional connection of each subsequence of samples is obtained, and the cluster center of all classes is formed into a regression matrix. Finally, the regression coefficient of the sample is calculated according to the regression matrix, and it is classified as a feature. The validity of multitask fusion Lasso method for dynamic FBN modeling is verified. MRI data set was used to verify the classification effect of multitask fusion Lasso model to construct dynamic FBN. ADNI) exposes the f MRI dataset to Alzheimer's patients. The classification performance of the algorithm was evaluated with accuracy, sensitivity and specificity in three groups of data from patients with early mild cognitive impairment and healthy subjects. The accuracy of this method is 92.31% and 84.00%, respectively. The experimental results show that this method is compared with the static FBN model and other traditional dynamic FBN models. Conclusion the proposed multi-task fusion Lasso method for constructing dynamic FBN can effectively preserve the sparsity of the network and the temporal smoothness of the sub-sequences. The multi-task fusion Lasso model can be used to construct dynamic FBN and mine the dynamic information of functional connection. At the same time, the whole algorithm can be used in the classification of brain diseases based on f MRI data.
【作者单位】： 中国科学院自动化研究所;中国中医科学院广安门医院;
【基金】：国家自然科学基金项目(61305018,61432008,61472423,61532006)~~
【分类号】：R445.2
【正文快照】： 第22卷/第7期/2017年7月王鑫,任燕双,张文生/动态功能脑网络模型的多任务融合Lasso方法brain region is the node,and a functional connectivity between each pair of brain regions is an edge.The functional con-nectivity between the brain regions can reveal disease，

本文编号：1487512