基于独立成分分析功能连接的抑郁症分类研究

Independent component analysis based functional connectivity for classification of depression

已有的功能连接研究大多根据脑图谱构建全脑功能连接,但目前可选用的脑图谱种类有限,且采用不同脑图谱的分析结果可能存在一定的差异。针对上述问题,利用独立成分分析方法研究了抑郁症辅助诊断问题。首先利用组独立成分分析提取独立成分并构建全脑功能连接网络,然后采用Boost FS(boosting feature selection)方法进行特征选择,最后应用多元模式分析方法对20名抑郁症患者和21名健康被试进行分类。实验分类准确率达到95.12%,错分了一名抑郁症患者和一名健康被试。进一步分析表明,具有较强分辨能力的脑网络为感觉运动网络、默认网络和视觉网络,与已有基于脑图谱的研究结果基本一致,从而说明了基于独立成分分析方法的合理性,使其可能成为抑郁症辅助诊断的一种新方法。

Previous functional connectivity studies usually utilize brain atlas to construct whole-brain functional connectivity network. However,available brain atlas is limited and the results using different brain atlas might be different. In order to overcome the problems above,this paper investigated the computer aided diagnosis of depression using independent component analysis method. Firstly,it extracted independent components to construct whole-brain functional connectivity network,and then used Boost FS method for feature selection. Finally,it employed multivariate pattern analysis method to classify 20 depressed patients and 21 healthy controls. The classification accuracy of the proposed method was up to 95. 12%（ only one healthy control subject and one depressed patient were classified incorrectly）. Further analysis demonstrates that the most discriminative brain networks are sensorimotor network,default mode network and visual network,which is consistent with the existing results of brain atlas. This indicates that the proposed independent component analysis based method is reasonable,and it might be a new method for the diagnosis of depression.