Dynamic effective connectivity network based on change points detection

Human brain networks can be modeled as a system of interconnected brain regions which are recorded by time-dependent observations with functional magnetic resonance imaging (fMRI). In order to spot trends, detect anomalies, and interpret the temporal dynamics, it is essential to understand the connections among distinct brain regions, and how these connections evolve over time. However, the change points of dynamic reorganization in brain connectivity are unknown, which may occur frequently during the scanning session. In this paper, we introduce a fused lasso regression approach to detect the number and position of rapid connectivity changes of subject and subsequently estimate the brain effective connectivity networks within each state phase lying between consecutive change points by conditional Granger causality method from fMRI time series data. The performance of the method is verified via numerical simulations and the obtained classification accuracy with support vector machine (SVM) was 86.24% in 140 subjects from Alzheimer's Disease Neuroimaging Initiative (ADNI). Compared with static EC model and conventional dynamic EC model based on sliding window technique, the experimental results show that the fused lasso achieved better classification effect, which probably due to better dynamic description. The result shows that the dynamic effective connectivity based on change points detected by fused lasso method is a better feature for classification.