Modeling the Link Between Structural Connectivity Network Disruption and Performance and Activity Limitation in Stroke

Abstract

Objective: To quantify the relationship between structural white matter (WM) connectivity network disruption due to ischemic stroke lesions and 18 clinical assessments of impairment and activity limitation. Design: Causation survey Setting: Inpatient rehabilitation unit (IRU) Participants: This original sample was 92 subjects with stroke admitted to the IRU. A total of 47 subjects (age: 71.2 12.3, NIHSS: 7.8 5.9) satisfied the inclusion criteria: 1) ischemic stroke 2) MRI scans acquired at the home institution and 3) clearly apparent hyper-intensities on diffusion weighted images. Interventions: N/A Main Outcome Measure(s): The Network Modification (NeMo) Tool quantified regional changes to the structural connectivity network in each patient by projecting WM abnormalities onto gray matter regions they connect. A Partial Least Squares Regression approach was then used to predict various cross-sectional outcome measures. Results: Model fits ranged from moderate to high (R: 0.28-0.85) for the 18 clinical assessments, but in all cases the PLSR models explained more variance in the outcome measure than stroke lesion volume alone. Significance testing of regression coefficients identified regions that were important for each outcome measure. Models of eloquent functions, e.g. motor and aphasia, provided validation of the methodology. Conclusions: This study demonstrates that the NeMo tool has the potential to be used to predict the impact of stroke size and location on impairment and activities limitation. This approach quantifies the classic lesion study, allowing exploration of the complex relationships between structural brain networks and human behavior. Machine learning tools like this one will play an increasing role in such investigations.