Research On Application Of Local Information Method In Brain Functional Network Modeling

In recent years, the brain network has aroused people’s wide attention, and has become a hot issue. The brain network is divided into the brain structural network and the brain functional network, and there is a very close relationship between the two parts of brain network. Studies of brain network are focused on the topological characteristics and analyzing the network metrics. The brain network is a very complex system, which is not only reflected in the billions of neurons in the brain network, but also in the different connection modes of behavior and cognition. Nowadays, the researches of the brain network are from different perspectives, including the neurons level, the voxel level and the level of regions. The regions level should be the relatively large grain sized research level, however, studies have shown that, the current computation capacity can’t meet the demand when the number of regions reaches 4000. The studies of brain networks, especially the functional brain networks depend on scientific calculations in great extent. And because of the low computation capacity, many scholars began to seek for the methods of applying simulation to study the topological metrics of brain functional network. Thus made the brain network modeling a very hot issue.With the continuous development of research on the brain network modeling, people have tried many features in the modeling of brain functional network, and has achieved many important results. After a large amounts of researches, people have found that, the structural degree shows great advantage in the forecast of functional connections. Also, there are some other researches based on the anatomical distances to stimulate the brain functional network, which have also achieved very good results. In addition, some others implied the common neighbor as the similarity indicator of the nodes to the brain functional modeling. And the result is proved to be wonderful. As to the mathematical model, the most popular model is the one that takes both the structural properties and functional properties into consideration. After a deep research, this paper also takes the mathematical model based on structural properties and functional properties of brain network.In this paper, the main innovational work are as follows:First, the local information method of link prediction is introduced into the brain functional network modeling. The local information indicators refer to forecasting whether there is a connection or not between a pair of nodes according the node similarity. A previous study has used the common neighbor as the node similarity indicator to stimulate the brain functional network, and common neighbor is a classical local information indicator. Except for common neighbor, there are also many local information indicators. So, weather the results can be as good as the common neighbor if we apply the rest local information indicators to the brain functional network modeling? In order to answer the question, this paper has chosen 8 local information indicators and applied them to the brain functional modeling. In the 8 indicators, some are calculated based on the node degree, some are calculated based on common neighbor, the rest ones are calculated based on both the common neighbor and the node degree. Common neighbor and the node degree are the most famous modeling factors. So, this paper is mainly study which kind of local information indicators are the best one to stimulate the brain functional network.Second, a new brain network similarity calculation method was proposed. To finish the modeling experiment is just the first step, the more important work is to evaluate the performance of the models. In the brain network field, scholars tend to apply the global and local metrics of brain network, such as the characteristic path length, the clustering coefficient, the global efficiency and the local efficiency to indicate the network performance. Global and local metrics can show the communication efficiency of brain network from the perspective of complex network and graph theory. However, there is no authoritative methods to indicate the brain network similarity. In this paper, we combined the two method above and proposed a new network similarity calculation method, which applied the relative error to assess the difference between the model network metrics and the real network metrics. In addition, we added all the relative errors together and made an index of E value to show the overall similarity between the model networks and the real networks. The difference becomes larger if the relative error goes larger. So, we took the style of E value from the previous search and made the relative error denominator. So, the similarity degree goes larger as the E values goes larger.Third, A prior estimate method based on the link prediction accuracy was proposed, which can predict the result of local information indicators being used in brain network modeling. Brain network modeling is a very complicated process, including data collection, data pretreatment, the brain functional network building, network metrics computing and the model networks evaluation. To complete the whole modeling process, we need to pay a lot of time cost and calculation cost. In this paper, we proposed a method based on the link prediction accuracy to foretell the result before stimulating the brain network. We hypothesized that, if the link prediction accuracy index Precision Power is larger, the local information indicator is more fit for the brain network modeling. To verify the hypothesis, we tested the correlation relationship between the E value and the PrecisionPower, and the result shows that, there is a significant linear relation between the E value and the Precision Power. So, we can judge the modeling result of using local information indicator in brain network modeling according the link prediction accuracy.