基于高阶最小生成树脑网络的多特征融合分类方法

现有的基于脑疾病的分类方法的研究使用的都是传统的低阶功能连接网络。低阶功能连接网络可能会忽略复杂的大脑区域之间动态的相互作用的模式。高阶功能连接网络能够反映网络中包含的丰富的动态时间信息,但原有的高阶功能连接网络使用聚类的方法降低了数据维度,使得构建的网络无法进行有效的神经学解释;其次,高阶功能连接网络由于规模较大,在利用复杂网络或图理论计算一些拓扑指标时消耗较大。基于此,提出了一种高阶最小生成树网络的构建方法,然后计算了传统的可量化网络指标(度和离心率)并结合频繁子图挖掘技术来挖掘具有判别能力的子网络,最后采用多核支持向量机进行分类。实验结果表明所提方法的分类精确度高达97.54%,获得了很好的分类性能。

Multi-feature Fusion Classification Method Based on High-order Minimum Spanning Tree Brain Network

Existing researches on classification of brain diseases is based on the traditional low-order functional connectivity network.Low-order functional connectivity network may overlook the complex and dynamic interaction patterns among brain regions,which are essentially time-varying.The high-order functional connectivity network can reflect the abundant dynamic time information contained in the network.However,the traditional high-order functional connectivity network adopts the clustering method to reduce the dimensionality of the data,making the construced network can not be effectively interpreted from the perspective of neurology.Even more importantly,due to the large scale of the high-order functional connectivity network,it is very time-comsuming to use some complex network or graph theory to calculate some topological properties.Therefore,this paper proposed a method for constructing a high-order minimum spanning tree network,calculated the traditional quantifiable network properties (degree and eccentricity),and used frequent subgraph mining technology to capture the discriminative subnetworks as features.Then,this paper applied a multi-kernel learning technique into the corresponding selected features to obtain the final classification results.The experimental results show that the classification accuracy is up to 97.54%.