Detection of differentially developed functional connectivity patterns in adolescents based on tensor discriminative analysis

Adolescence is a crucial time for the development of functional connectivity, and the tuning process changes under different physical and cognitive status. However, there is still a lack of knowledge on where and how functional development changes across this age range. In this paper, we introduce a discriminative tensor decomposition method to detect differentially developed functional connectivity patterns. The method is a combination of the Fisher's discriminative analysis and CANDECOMP/PARAFAC(CP) decomposition, linked by a novel discriminative developmental ratio. A sequential orthogonal decomposition algorithm is then proposed to efficiently solve the problem. The method is validated by a real dataset, which contains functional connectivity maps for each individual. We separate the subjects by sex and use a sliding window approach over age to obtain two correlation tensors. Applying our method to the two tensors, we can detect connectivity patterns with differential development in sex. Our results show evidence for developmental differences between girls and boys that are largest during puberty.