UC Irvine

Stroke is highly heterogeneous, with patients demonstrating variation in infarct characteristics, baseline impairment, degree of spontaneous recovery, and response to treatment. There is ample literature on the neural correlates of post-stroke variation. However previous methods incompletely characterize inter-individual differences and have limitations for clinical adoption. The studies herein examine the use of dense-array electroencephalography (EEG) measures of brain function for predicting response to motor training in non-stroke control and stroke populations. In Chapter 2, a partial least squares regression (PLS) model found resting-state connectivity was a robust predictor of subsequent response to motor training (R2 = 0.81) in non-stroke individuals. Results in Chapter 2 were confirmed in Chapter 3, and extended when the same resting-state connectivity measure was shown to demonstrate specificity with respect to the content of motor training. Specifically, individuals with increased connectivity between electrodes overlying left primary motor cortex (M1) and left premotor cortex (PM) showed greater improvement with training of the motor sequence learning task, but lesser improvement with training of the visuospatial learning task. The same EEG methods were applied to a chronic stroke population in Chapter 4. Separate PLS models found baseline resting-state connectivity was related to baseline impairment (R2 = 0.78) and predicted motor gains across the 4 week arm motor therapy (R2 = 0.79). Finally, in Chapter 5, similar EEG methods were used to identify the neural correlates of impairment in acute stroke (0 - 12 days). A PLS model showed relative delta power across the brain accounted for a substantial proportion of variance in early post-stroke impairment state (R2 = 0.72). Furthermore, the neural correlates of acute stroke impairment were found to vary according to infarct size subgroup; the model for small strokes was driven by ipsilesional signals, whereas the model for large strokes was driven by contralesional signals. In sum, the current studies demonstrate that PLS modeling EEG measures of resting brain function provide useful insights into basal differences in brain state which can be used to predict the capacity of an individual brain to undergo plasticity and thus respond to motor training in non-stroke and stroke populations.