Stroke remains a leading cause of adult disability, underscoring why research continues to focus on advancing new treatment methods and neurophysiological indexes. While these studies may be effective, many lack a clear theoretical framework. The current study first determined the optimal combination effects of mirror therapy (MT) with transcranial direct current stimulation (tDCS) on the premotor or primary motor cortex on its short-term and sustained clinical outcomes. We then introduced electroencephalogram (EEG) indexes derived from the gating-by-inhibition model to explore the underlying therapeutic mechanisms. The EEG indexes used in this study focused on the functional involvement for motor generation: alpha power at temporal regions (inhibiting non-motor activity) and central-frontal regions (releasing motor regions from inhibition). Results showed that post-training benefits, measured by Fugl-Meyer Assessment (FMA), were similar across 3 tDCS interventions (premotor, primary motor, sham). EEG seemed more sensitive to the training, with notable responses in the premotor tDCS group. Three months after training, only the premotor tDCS group maintained the gains in FMA, with these improvements correlated with the EEG indexes. Again, this pattern was specific to premotor tDCS. Since the gating-by-inhibition model suggests that EEG index reflects an individual's psychomotor efficiency, we also found that the baseline EEG index could predict FMA retention. Our findings demonstrate the superiority of combined premotor tDCS with MT and identify functionally oscillatory alpha-band activity in the temporal and central-frontal regions as potentially underlying the therapeutic mechanism. An individual's spatial pattern of EEG may be effective in predicting upper extremity retention effect.