Brain Network Classification Based On Graph Mining With Applications

In recent years, with the development of medical imaging techniques, discovering biomark-ers and connectivity characteristics which are sensitive to brain disease from neuroimaging data, and using it for brain disease classification, has become a new focus of research. It has been a research trend to analyze brain network and obtain rules to predict unknown data based on data mining and machine learning techniques. Based on the graph mining techniques, we did some research on brain network classification. The main innovation and research work are summa-rized as follows:Firstly, we proposed a frequent and discriminative subnetwork mining with single network based brain network classification method. Our hypothesis is that there exist common subnet-works in the same group, and different subnetworks in different groups. The different subnet-works between different groups are utilized for brain network classification. Specifically, we firstly extract subnetworks in different groups with using frequent subgraph mining technique. Then, we measure the discriminative ability of those frequent subnetworks using the proposed discriminative subnetworks selection algorithm and select the most discriminative subnetworks for subsequent classification. The experimental results show that the proposed method can not only significantly improve the classification performance, but also can detect potential functional or anatomical connectivity characteristics and brain regions which are sensitive to brain disease.Secondly, we proposed a discriminative subnetwork mining with multiple network fusion based on the single network method. Our hypothesis is that different thresholded connectivity network can generate different connectivity network, which can reflect different levels of con-nectivity topological characteristics. Our aim is to utilize those connectivity topological charac-teristics for brain network classification. Specifically, we firstly use different threshold to thresh-old connectivity network. Then, we perform frequent and discriminative subnetwork mining on network group, respectively. Finally, we combine all the selected discriminative subnetwork-s for subsequent classification. The experimental results show that multiple networks fusion based frequent and discriminative subnetwork mining can obtain more stable classification per-formance, and can obtain more discoveries about brain connectivity topological characteristics.