Advancing brain network models to reconcile functional neuroimaging and clinical research

Functional magnetic resonance imaging (fMRI) captures information on brain function beyond anatomical alterations traditionally visible to neuroradiologists. However, the fMRI signal is complex and noisy, and so far fMRI has brought limited value in a clinical research context. We argue that solutions can be found in richer fMRI-based models such as statistical, biophysical and decoding models. These models extract clinically relevant information regarding biological mechanisms and features for classification and prediction (interpretability). Moreover, they are suitable to directly predict clinical variables from their parameters (predictability). We give guidelines for useful applications and pitfalls of such fMRI-based models in a clinical research context, and we look beyond currently used models. In particular, we provide arguments that clinical relevance of fMRI calls for better fMRI activity models incorporating both interpretability and predictability. We illustrate how this synergy of interpretability and predictability can be achieved by combining biophysical models with decoding models. These hybrid models entail reliable and biologically meaningful model parameters. In our view, this synergy is fundamental for the discovery of new pharmacological and interventional targets and the use of models as biomarkers.