论文部分内容阅读
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。
Purpose The functional brain network (FBN) study of traditional resting functional magnetic resonance imaging (f MRI) is based on the assumption that the FBN remains constant throughout the scan. However, recent research shows that FBN is dynamic and contains a wealth of information. In this paper, we propose a multi-tasking fusion of the minimum absolute value shrinkage and selection operator (Lasso) method to construct a dynamic FBN for resting state f MRI. The multi-task fusion Lasso method proposed in this paper can preserve the sparseness of the network and the temporal smoothness of sub-sequences when building dynamic FBN. Specifically, the overlapping resting MRI sequences were obtained by sliding window method firstly. Then the multi-task fusion Lasso method was used to jointly estimate the functional connections of all the subsequences of a sample so as to construct a dynamic FBN. Using k-means clustering Then the clustering centers of all classes are formed into a regression matrix. Finally, the regression coefficients of the samples are obtained according to the regression matrix, which is classified as a feature to verify the multi-task fusion Lasso method Validity of dynamic FBN modeling. Results The published f MRI data set was used to validate the classification of dynamic FBN by multi-task fusion Lasso model. The experiment used three sets of data from patients with Alzheimer’s disease, patients with early mild cognitive impairment and healthy subjects in the f MRI data set published by the Alzheimer’s Disease Neuroimaging Program (ADNI), and used the accuracy, sensitivity and Specificity is used to evaluate the classification performance of the algorithm. In three groups of two classification experiments, the accuracy of this method reached 92.31%, 80.00% and 84.00% respectively. The experimental results show that the proposed method can achieve better classification results than the static FBN model and other traditional dynamic FBN models. Conclusion The proposed multi-task fusion Lasso method to construct a dynamic FBN can effectively preserve the sparseness of the network and the time smoothness of sub-sequences, and improve the classification effect of the algorithm, to some extent, to help the diagnosis of brain diseases. The multi-task fusion Lasso model can be used to construct dynamic FBN, to mine the dynamic information of functional connections, and at the same time the whole algorithm can be used in the classification of brain diseases based on f MRI data.