Abstract: Resting-state functional brain networks have been widely studied in brain disease research. However, traditional functional connectivity networks mainly focus on the certain graph, and ignore uncertain information between brain regions. Therefore, the uncertain brain network is analyzed, which does not need to select threshold, and can more accurately construct model of the functional connection network. At the same time, this paper applies frequent subgraph mining to the uncertain graph, and proposes several new discriminative feature selection methods. The classification results show that the fMRI classification method based on uncertain brain network effectively improves the diagnostic accuracy of depression.

Key words: depression, uncertain brain network, frequent subgraph mining, discriminant feature selection

摘要： 静息态功能脑网络在脑疾病研究中得到了广泛的应用。然而传统的功能连接网络分析主要集中在确定图上，忽视了大脑区域之间的不确定信息。基于此，对不确定脑网络进行了研究，该方法不需要进行阈值选择，而且可以更准确地对功能连接网络进行建模。同时，将频繁子图挖掘应用到了不确定图上，并提出了几种新的判别性特征选择方法。分类结果显示，基于不确定脑网络的磁共振影像分类方法有效地提高了抑郁症诊断的准确率。

关键词: 抑郁症, 不确定脑网络, 频繁子图挖掘, 判别性特征选择