Introduction.

Previous studies have shown that active professional fighters who experience repeated head trauma may suffer from cognitive impairment when compared to age matched healthy controls^1,2. Previous neuroimaging studies have identified abnormalities in brain regional volumes, cortical thickness and structural connectivity in cognitively impaired active fighters, which correlate with their cognitive task performances^3,4. However, functional connectivity (FC) changes in active fighters remain unclear and dynamic functional connectivity abnormalities in active fighters have not yet been investigated. Here we explored both static and dynamic functional connectivity differences between cognitively impaired and non-impaired active fighters using resting-state fMRI.

Methods.

Subjects. A total of 252 active professional fighters were recruited at our center as part of the Professional Fighters Brain Health Study². Each subject went through neuropsychological assessments to measure psychomotor speed (PSY) and processing speed (P). 68 subjects were classified as cognitively impaired based on their PSY and P scores being 1.5 standard deviation below the average⁵ (65Males, age=29.80±6.20 years, years of education (YOE)=13.03±2.12 years). 65 matched non-impaired fighters (58Males, age=28.78±5.27 years, YOE=13.28±1.63 years) were selected and also included in the analysis. Resting-state fMRI data were collected for all subjects on a 3T Siemens scanner (TR/TE/resolution=2.8s/28ms/2x2x4mm³, 30 slices, axial acquisition, 137 time frames). Static functional connectivity analysis. After standard preprocessing steps, fMRI data from all subjects were normalized to the standard MNI-152 2mm template. Functional brain networks were obtained through group independent component analysis (ICA). Out of 100 ICA components, 48 components were identified as resting-state networks. Subject-specific spatial maps and time-courses for each network were reconstructed using GIG-ICA⁶. Spatial maps of each network were compared between impaired and non-impaired fighters. Functional connectivity among different networks (FC matrices) were computed using the correlation of the subject-specific time-courses between each network pairs. FC matrices were finally compared between impaired and non-impaired fighters. Dynamic functional connectivity analysis. Dynamic FC was estimated using a sliding-window method⁷, with the optimum window-size computed through empirical mode decomposition method⁸. Dynamic FC matrices, i.e. correlation matrices among different networks, were computed within each window and a k-means clustering was further carried out to compute dynamic FC states. Number of dynamic states was determined with the elbow criteria. Subject-specific dynamic FC matrices for each state were further computed by averaging the FC matrices of all windows clustered to the state. The dynamic FC matrices were also compared between impaired and non-impaired fighters for each state. Classification. Automated feature selections (using LASSO⁹) and classifications (using radial basis function classifier)¹⁰ were finally carried out using static FC matrices and dynamic FC matrices of all states as input features, separately.

Results.

Fig.1 describes the flowchart used in the analysis. Static network comparison. In the anterior default mode network (DMN) and cerebellum network, significant decreased network strengths are observed in impaired fighters (p<0.05, FDR corrected), with the cluster center located at the superior frontal gyrus and cerebellum crust1, respectively (Table. 1). The reduced network strength in anterior DMN is negatively correlated with the number of fights in all fighters (p=0.03, Fig. 1), adjusted for the age, gender and YOE. Static FC comparison. No significant between network difference is observed in the static FC comparison at FDR corrected p<0.05. Dynamic FC comparison. Four dynamic functional states were identified (Fig. 2(A)) and impaired fighters spent more time (but not significant, Fig 2(B)) in a weakly connected state 2. In this state (state 2), impaired fighters showed reduced functional connections between cognitive control networks and motor networks (p<0.001), and increased functional connections between cerebellum networks and motor networks (p<0.001), as compared to non-impaired fighters. Classification. Table 2 lists the selected features using LASSO and the accuracy of classifying impaired and non-impaired fighters in the independent testing set. The accuracy was 0.65 when using static FC matrices as features and 0.77 when using dynamic FC matrices as features.

Discussions and conclusions.

The higher classification accuracy with dynamic FC as features suggests the time-varying brain activities carry richer cognitive impairment-related information. This study is the first attempt to investigate abnormalities in both static and dynamic FC of cognitively impaired active fighters, which provides evidence of both disrupted static functional networks and altered dynamic functional connections in impaired active professional fighters.

Acknowledgements

The study is supported by the National Institutes of Health (grant number 1R01EB014284 and P20GM109025), grants from Lincy Foundation, the Peter and Angela Dal Pezzo funds and the young scientist award at CCLRCBH.