抑郁症复杂脑网络社团结构差异分析及分类研究

摘要

为比较抑郁症脑网络结构差异及实现患者自动识别,提出了基于功能脑网络社团结构特征的机器学习方法.利用静息态功能磁共振影像数据构建功能脑网络,利用基于“堆结构”的贪婪算法进行社团划分,从脑网络模块结构的角度分析正常人和抑郁症患者的差异,并将脑网络的模块指标用于机器学习方法.利用统计显著性为阈值以筛选特征,以判断不同特征数目对分类模型的性能影响.实验结果表明,神经网络算法在28个特征下(P＜0.05),分类正确率最高达90.50％.%In order to identify patients automatically,a machine learning method of functional brain network community structure characteristics is put forward to compare the brain networks between the major depression disorder patients and normal controls.Resting state functional brain networks are constructed and greedy algorithm is used to divide community structure.From the point of view of the modularity of the brain network,the differences of modularity metrics of the brain network between normal controls and patients are found and are used as classification features into machine learning method.Statistical significance is used as the threshold for selecting features in order to measure the accuracies with different number of features.Results show that neural network algorithm exhibit the highest average accuracy (90.50％) with 28 features (P ＜ 0.05).