Structure-function dependencies as informative features for brain decoding and fingerprinting

Functional magnetic resonance imaging has proven useful to decode task-specific brain activity and act as brain fingerprints, that is, allowing accurate identification of individuals within a large group. In this context, the relationship between functional activity and the underlying structural wiring, extracted from diffusion-weighted magnetic resonance imaging, can be a new imaging-based biomarker characterizing tasks and individuals. To investigate this, we used a recent graph signal processing framework to quantify the regional function-structure dependencies through the structural decoupling index (SDI), in 100 unrelated healthy volunteers from the Human Connectome Project, scanned both during resting-state and seven different tasks. SDI values of different tasks and subjects could be leveraged for accurate classification using linear discriminant analysis. The accuracies were very high and better than the ones obtained with functional connectivity alone. Further, we could identify two distinct brain networks, including most discriminative regions for task decoding and fingerprinting, respectively. These results show that rich information is contained in brain function-structure relationships, and that these provide new promising signatures of tasks and subjects.