No abstract

Visual Abstract

Large neuroimaging datasets, including information about structural (SC) and functional connectivity (FC), play an increasingly important role in clinical research, where they guide the design of algorithms for automated stratification, diagnosis or prediction. A major obstacle is, however, the problem of missing features (e.g., lack of concurrent DTI SC and resting-state fMRI FC measurements for many of the subjects).We propose here to address the missing connectivity features problem by introducing strategies based on computational whole-brain network modeling. Using the ADNI dataset for proof-of-concept, we demonstrate the feasibility of virtual data completion (i.e., inferring "virtual FC" from empirical SC or "virtual SC" from empirical FC), by using self-consistent mean-field network simulations and analytic approaches. Furthermore, we use similar procedures to perform dataset augmentation, i.e., complementing the original dataset with a large number of realistic surrogate connectivity matrices. We thus show that algorithms trained on virtual SCs and/or FCs can achieve performance in the unsupervised classification of control subjects and patients comparable to when trained on actual empirical data. Furthermore, the combination of empirical with virtual data allows algorithms to learn better how to extract information relevant for discrimination, resulting ultimately in superior classification performance.