Abstract: In the study of modularization attributes of dynamic brain function network, Louvain algorithm maximizes the modularity value, which leads to low recognition of dynamic brain function network modules. For solving this problem, a time-constrained Louvain algorithm is proposed. In this algorithm, the iterative end condition for time constraint model is constructed based on the distribution of modularity values on the entire data acquisition time period. With this time constraint condition, the ba- lance of both scale and quantity of modules can be achieved, which will be helpful to ensure the reasonability of modules. When this method is used in the module partitioning experiment of resting-state brain function network, the experiment results show that, compared with Louvain algorithm, the time- constrained Louvain algorithm can provide a more reasonable modularity recognition result. Furthermore, the dynamic brain function network modularity comparison experiments with this method between healthy people and autistic patients are carried out, and the results prove that there are significant differences in the modularization features between these two groups, which can verify the effectiveness of this method. 

Key words: dynamic functional connection, modularization, time-constrained condition, Louvain algorithm, resting state fMRI