Introduction

Concussion or mild traumatic brain injury (mTBI) remains a critical and poorly understood health issue, especially for young athletes participating in contact sports. While MRI studies have reported variable changes post-concussion, longitudinal multi-modal studies are required to gain a more complete understanding of the brain’s response and the risk of further concussions or more serious neurological disorders like second impact syndrome and chronic traumatic encephalopathy (CTE)^1,2. With increasing study design complexities, comparing different MRI measures becomes non-trivial. Linked independent component analysis (LICA) is a data-driven approach that can elegantly fuse different modalities to understand biologically relevant modes of shared inter-subject variance³. Here we examine female varsity rugby players using multi-parametric MRI longitudinally after a diagnosed concussion and compare to their non-concussed team mates over a period of 5 years to examine (a) acute and (b) persistent brain alterations after a concussion, and (c) the relationship of any changes to subject-specific clinical measures or concussion history.

Methods

Diffusion tensor imaging (DTI) with 64 gradient directions and a 10-minute resting state functional MRI (RS-fMRI) scan were acquired on a 3T MRI (Siemens) in non-concussed rugby players throughout the in- and off-season and compared to concussed teammates (n = 21) acutely (24-72 hours) and longitudinally (3- and 6-months) after a diagnosed mTBI. Turbo spin echo high-resolution anatomical images were used to confirm no hematological pathology indicative of more serious TBI. Diffusion data was eddy current corrected and analyzed using non-linearly registered tract-based spatial statistics to create FA-skeletonized maps (Figure 1a) of fractional anisotropy (FA), mean (MD) and axial (AD) diffusion⁴. RS-fMRI data was preprocessed in FSL and denoised using single-subject ICA⁵. Group ICA and dual regression algorithms were applied to temporally concatenated cleaned data to identify average resting state networks (RSN) and create subject-specific RSN maps, respectively, including the default mode network (DMN), cerebellar network, and the executive network (Figure 1b)⁶. DTI and RS-fMRI maps were concatenated by subject and analyzed using LICA with a modest model order (i = 15) given the longitudinal nature of this dataset. Components related to the subject’s age at the time of the scan or dominated by a single-subject (>10% variance) were not considered. The analysis was repeated for +/- 1 model orders to confirm robustness of relevant components that had associated subject weightings with a significant main effect for group using an AVONA with Tukey post-hoc correction or significantly correlated with clinical measures (Sports Concussion Assessment Tool [SCAT] scores) or concussion history (Bonferroni corrected, p < 0.05).

Results

We identified two components that had a significant main effect for group (p < 0.001). Component 4 (Figure 2) was heavily influenced by diffusion metrics (Figure 1c) and off-season subject weights were significantly higher than in-season (p < 0.05) or acute post-concussion data (p < 0.001), generally suggesting decreased MD, AD within the corpus callosum, inferior deep white matter structures and corticospinal tracts after concussion and possibly even sub-concussive impacts (Figure 3a). Component 5 had contributions from all 6 MRI metrics (Figure 4) and had significantly higher subject weights for 3- (p < 0.005) and 6-months (p < 0.0005) post-concussion compared to in- and off-season data (Figure 5a). Furthermore, when looking at only non-concussed athletes, we observed significantly higher component weights for subjects with any concussion history compared to those without (p < 0.05) that also correlated significantly with the total number of concussions (r = 0.51, p < 0.01 x 10^-10). This indicated decreases in MD and AD within the body and splenium of the corpus callosum, fornix and thalamus with some spatially related changes in functional connectivity beyond symptomatic recovery after mTBI. We investigated component 4 and 5 subject weightings longitudinally in concussed athletes using repeated measures ANOVA (n = 9) and found consistent changes in each individual player (Figure 3b and 4b).

Discussion

In this longitudinal multi-parametric MRI study of concussed and non-concussed female rugby players we identified linked components that described both acute and persistent brain changes after a concussion that also related to concussion history. We also found significant effects between in- and off-season data in the absence of an mTBI that warrants further investigation in order to isolate the effects of subconcussive impacts on the brain. These data-driven techniques were able to fuse functional connectivity and diffusion microstructure changes. Significant components were robust across model orders and within individual subjects longitudinally. MRI changes persisted beyond symptomatic recovery and a physician’s clearance to return to play. The increased risks of multiple concussions or second impact syndrome remain to be elucidated, and future research efforts need to follow athletes further to understand how or if these longitudinal changes recover or manifest into neurodegenerative processes.

Acknowledgements

The authors have no conflicts of interest to disclose and would like to thank and commend the players and their parents/guardians for their willingness to participate in this study. We would like to acknowledge the excellent support provided by the coaches and MRI technicians.

References

1. Bigler ED, Maxwell WL. Neuropathology of mild traumatic brain injury: Relationship to neuroimaging findings. Brain Imaging Behav. 2012; 6:108–136. 2. Daneshvar DH, Riley DO, Nowinski CJ, et al. Long-term consequences: effects on normal development profile after concussion. Phys Med Rehabil Clin N Am. 2011; 22: 683–700. 3. Groves AR, Beckmann CF, Smith SM, et al. Linked independent component analysis for multimodal data fusion. Neuroimage. 2011; 54: 2198–2217. 4. Smith SM, Jenkinson M, Johansen-Berg H. Tract-based spatial statistics: Voxelwise analysis of multi-subject diffusion data. NeuroImage. 2006; 31:1487-1505. 5. Griffanti L, Douaud G, Bijsterbosch J, et al. Hand classification of fMRI ICA noise components. Neuroimage. 2017; 154: 188–205. 6. Smith SM, Fox PT, Miller KL, et al. Correspondence of the brain’s functional architecture during activation and rest. Proc. Natl. Acad. Sci. 2009; 106: 13040–13045.