Clustered subsampling for clinically informed diagnostic brain mapping

Brain based diagnostic systems have recently received attention as a tool in the characterization and diagnosis of a variety of neurodevelopmental and psychiatric disorders. Nonetheless, a majority of disorders are still diagnosed entirely based on symptom assessments and behavioral correlates. We therefore propose a method for fusing brain responses with clinical measures for improved diagnosis. To this end, we utilized the flexibility of clustered random subspace brain mapping to detect regions where brain responses in conjunction with a clinical measure could reliably differentiate patients from control subjects. We demonstrate the approach on realistically simulated functional magnetic resonance imaging (fMRI) brain activity and a clinical parameter. We show that the method efficiently identifies brain regions where fused analysis of brain responses and clinical parameters improves diagnosis compared to either measure alone. The proposed method is easy to implement and highly flexible, offering an appealing basis for multimodal brain mapping.