Background: Postconcussion syndrome (PCS) or persistent symptoms of concussion refers to a constellation of symptoms that persist for weeks and months after a concussion. To better capture the heterogeneity of the symptoms of patients with PCS, we aimed to separate patients into clinical subtypes based on brain connectivity changes. Methods: Subject-specific structural and functional connectomes were created based on diffusion weighted and resting state functional magnetic resonance imaging, respectively. Following an informed dimensionality reduction, a Gaussian mixture model was used on patient-specific structural and functional connectivity matrices to find potential patient clusters. For validation, the resulting patient subtypes were compared in terms of cognitive, neuropsychiatric, and postconcussive symptom differences. Results: Multimodal analyses of brain connectivity were predictive of behavioral outcomes. Our modeling revealed two patient subtypes: mild and severe. The severe subgroup showed significantly higher levels of depression, anxiety, aggression, and a greater number of symptoms than the mild patient subgroup. Conclusion: This study suggests that structural and functional connectivity changes together can help us better understand the symptom severity and neuropsychiatric profiles of patients with PCS. This work allows us to move toward precision medicine in concussions and provides a novel machine learning approach that can be applicable to other heterogeneous conditions.
Keywords: clustering; connectivity; machine learning; magnetic resonance imaging; postconcussion syndrome.