The study of structure-function relationships in the brain has been an active area of research in neuroscience. The availability of brain imaging data that captures the structural connectivity and functional co-activation of the brain regions has led to the study of multimodal technical frameworks that can help disentangle the mechanisms linking cognitive abilities and brain structural alterations. This paper analyzes the diffusion and resting state functional magnetic resonance imaging (dMRI and rs-fMRI) data collected from a population consisting of former athletes with a history of multiple concussions and healthy controls with no reported history of concussion. For each subject, the structural connectome is represented by a graph with its nodes associated with cortical brain regions and the adjacency matrix derived from dMRI. Each brain region is associated with a blood oxygen level dependent (BOLD) signal derived from fMRI. This paper uses the tools from graph signal processing (GSP) to select the brain regions of interest (ROIs) that have significant statistical differences in the extracted high and low graph frequency components of the region specific BOLD signal across former athletes and healthy controls, where the graph frequencies represent the extent of spatial variations of the BOLD signal across the brain. The selected ROIs have also been previously identified to be affected in the existing clinical studies on traumatic brain injuries (TBI). Furthermore, the dynamic functional connectivity profiles of the selected ROIs are determined by leveraging the high and low graph frequency components of the BOLD signal and a sliding window based approach. Interestingly, the graph frequency functional connectivity profiles reveal unique characteristics that are not apparent in the unimodal dynamic functional connectivity profiles based on fMRI. Our analysis reveals statistically significant differences in the dwell times in multiple dynamic graph frequency functional connectivity states for the two groups of subjects. Therefore, the results presented in this paper underline the significance of graph signal processing tools for multimodal analysis of brain imaging data and also provide promising direction for applications in clinical research and medical diagnosis.