Neuromark Fusion: A Replicable Multimodal Template for Structure-Function Fusion of Brain MRI

Abstract

Multimodal data fusion is a powerful technique for extracting shared and complementary information about the brain that is captured across neuroimaging modalities. Independent component analysis (ICA)-based approaches are among the most widely utilized methods for multimodal fusion, as they are data-driven, robust to noise, and capable of identifying complex, hidden linkages of varying strengths across high-dimensional datasets. However, the data-driven nature of ICA fusion approaches can make comparisons across analyses difficult without a normative framework in place. In this work, we utilize resting state functional MRI (rsfMRI) and structural MRI (sMRI) scans from >15,000 subjects to generate a normative model of multimodal structure-function linkages that can be used as a template to guide ICA fusions of new datasets. When applying this template in two datasets, resultant components exhibit high correspondence to the template even in small sample sizes, and subject-level loadings from template-derived ICs show significant associations to age.