The training sessions will consist of high repetitions of finger flexion/extension and grip/object manipulation tasks, wrist flexion/extension and radial/ulnar deviation and forearm supination/pronation tasks performed while seated, standing or lying comfortably. All activities are typical of those conducted in physical therapy post-stroke except for the high number of repetitions. Some example tasks might include using a mallet to hammer pegs into a hole, pouring water from one cup into another, or spinning a pencil with the fingertips. Difficulty of tasks will be scaled to the capability level of each individual so that at least 200 repetitions of each exercise can be completed with 50-85% accuracy and a patient-reported difficulty of 5-9 out of 10.

The training sessions will consist of high repetitions of finger flexion/extension and grip/object manipulation tasks, wrist flexion/extension and radial/ulnar deviation and forearm supination/pronation tasks performed while seated, standing or lying comfortably. All activities are typical of those conducted in physical therapy post-stroke except for the high number of repetitions. Some example tasks might include using a mallet to hammer pegs into a hole, pouring water from one cup into another, or spinning a pencil with the fingertips. Difficulty of tasks will be scaled to the capability level of each individual so that at least 200 repetitions of each exercise can be completed with 50-85% accuracy and a patient-reported difficulty of 5-9 out of 10.

The primary outcome is the goodness of fit (R^2) of a regression model relating brain connectivity measures (predictors) to gains in motor function after stroke (clinical measurement).

Predictors: resting-state functional connectivity measured using resting-state fMRI (rs-fMRI). Connectivity is quantified as Fisher-transformed correlation coefficients (z scores, unitless) of signal measured in pairs of regions (edges) within the cortico-thalamic-cerebellar network. Predictors include baseline (pre-task) connectivity and connectivity changes calculated as the difference between post-task and pre-task rs-fMRI measurements.

Clinical measurement: change in Upper Extremity Fugl-Meyer Assessment (FM-UE) score (units: points), calculated as post-intervention FM-UE minus pre-intervention FM-UE.

Outcome: model goodness of fit (R^2, unitless). We use a custom LASSO-based variable selection method to construct the final model given multiple predictors and quantify the model fit via R^2.

The secondary outcome is the goodness of fit (R^2) of a linear regression model relating corticospinal tract injury (predictor) to gains in motor function after stroke (clinical measurement).

Predictor: percent corticospinal tract injury (%CST injury) quantified using a structural MRI brain scan (FLAIR). The FLAIR image shows the location and volume of lesioned tissue, and %CST injury is quantified as the percentage of CST axon bundles (derived from a previously published healthy template) that overlap with the lesion (predictor units: percent damaged axon bundles).

Clinical measurement: change in Upper Extremity Fugl-Meyer Assessment (FM-UE) score (units: points), calculated as post-intervention FM-UE minus pre-intervention FM-UE.

Outcome: model goodness of fit (R^2, unitless). We use linear regression to relate %CST injury to the change in FM-UE and quantify the model's fit via R^2.