Prediction-Identification Landscape for Brain Structure and Connectivity
Date:
Neuroimaging-derived brain phenotypes can be used to identify individuals and predict behavior. We investigated the prediction and identification performance of brain structure and connectivity phenotypes derived from structural, functional and diffusion MRI. Five behavioral domains were predicted: cognition, illicit substance use, tobacco use, personality-emotion traits, and mental health. We arranged the phenotypes on a two-dimensional prediction-identification landscape. Functional connectivity performed better at prediction than identification, whereas the converse was found for curvature and other structural phenotypes. Structural connectivity performed well for both tasks. High-resolution measures outperformed atlas-based counterparts. Our work can aid brain phenotype selection in future neuroimaging studies.