Abstracts

Rationale: Approximately 25% of traumatic brain injury (TBI) patients develop post-traumatic epilepsy (PTE). Studies have attempted to identify neuroimaging biomarkers of PTE in structural and functional MRI, however these modalities have limitations when analyzed independently. The presence of brain contusions on structural images is a major risk factor for developing PTE, but their impact on brain networks and function is not well characterized. Here, we present an approach to identify biomarkers through multimodal imaging and clinical data analyses that incorporate the presence of contusions.

Methods: This study investigated a cohort of 24 healthy control (HC), 20 TBI, and 16 PTE subjects who underwent neurocognitive tests, structural, and resting state functional MRI. Neurocognitive tests included BVMT, NAB, Trails, Stroop, HVLT, and SDMT. Images were co-registered to the MNI atlas and functional connectivity (FC) was computed based on the 100 ROI parcellation Yeo atlas corresponding to 7 resting state networks: Visual, Somato-Motor, Dorsal-Attention, Ventral-Attention, Limbic, Fronto-Parietal, and Default. A Random Forest was trained to assess the importance of demographics, neurocognitive scores, and presence of contusions in the classification of subjects into No-Seizure (HC and TBI) and Seizure (PTE) groups. The number of ROI affected by contusions in each network was correlated with the neurocognitive outcomes for the entire cohort. To evaluate FC differences, t-tests were performed to identify significant connectivity disruptions among the ROI of the 7 networks.

Results: The Random Forest model demonstrated an AUC of 0.84 ± 0.07 with BVMT, NAB-Dots, SDMT, and Trails as the features of highest importance in 5-fold cross validation (Fig. 1). In comparison, demographic features and contusions were relatively not important in classifying seizure state. The most significant correlations were among contusions of the Visual, Somato-Motor, and Default mode networks and neurocognitive tests associated with visual scanning, executive processing, task switching, verbal learning, and working memory (Fig. 2A). Seed based analysis identified 301 significant FC disruptions in ROI based on seizure group. The highest number of significantly disturbed network connections was also between the Visual and Somato-Motor and within the Default networks (Fig. 2B). The significance of these 3 networks in both structural and functional imaging and neurocognitive outcomes suggests a multimodal relationship.

Conclusions: Contusions in the Visual, Somato-Motor, and Default networks were most significantly correlated with neurocognitive outcomes – these 3 networks also demonstrated the highest degree of functional connectivity disruption when comparing seizure state. Our results suggest that contusions and connectivity changes in these networks may underlie maladaptive recovery processes that precipitate PTE and poor neurocognitive outcomes post-TBI. Leveraging a multimodal approach to identify biomarkers yields a more nuanced story of PTE risk than isolated analyses.

Funding: Department of Defense and Congressionally Directed Medical Research Program Epilepsy Research Program Idea Development Award: W81XWH-18-1-0655