The Neurophysiological Impact of Subacute Stroke: Changes in Cortical Oscillations Evoked by Bimanual Finger Movement

Abstract

Introduction. To design more effective interventions, such as neurostimulation, for stroke rehabilitation, there is a need to understand early physiological changes that take place that may be relevant for clinical monitoring. We aimed to study changes in neurophysiology following recent ischemic stroke, both at rest and with motor planning and execution. Materials and Methods. We included 10 poststroke patients, between 7 and 10 days after stroke, and 20 age-matched controls to assess changes in cortical motor output via transcranial magnetic stimulation and in dynamics of oscillations, as recorded using electroencephalography (EEG). Results. We found significant differences in cortical oscillatory patterns comparing stroke patients with healthy participants, particularly in the beta rhythm during motor planning ( p = 0.011 ) and execution ( p = 0.004 ) of a complex movement with fingers from both hands simultaneously. Discussion. The stroke lesion induced a decrease in event-related desynchronization in patients, in comparison to controls, providing evidence for decreased disinhibition. Conclusions. After a stroke lesion, the dynamics of cortical oscillations is changed, with an increasing neural beta synchronization in the course of motor preparation and performance of complex bimanual finger tasks. The observed patterns may provide a potential functional measure that could be used to monitor and design interventional approaches in subacute stages.

Transferring a deep learning model from healthy subjects to stroke patients in a motor imagery brain–computer interface

“…Figure 6 depicts these findings in a line graph. Although healthy individuals and stroke patients may exhibit variations in neural oscillations during MI [64,65,80], we demonstrated that through adaptation, one can obtain a comparable level of MI classification accuracy using the H-to-S transfer model as with the S-to-S transfer model for most patients. We observed that in few patients, the accuracy obtained using the adapted H-to-S model was better compared to that of the adapted S-to-S model, and vice versa.…”

“…To mitigate this issue, domain adaptation based transfer learning methods are commonly utilized [29,53]. In this study, the differences in MI-related neurophysiological patterns between the healthy and patient groups of subjects exacerbate the inter-subject disparities [64,65]. To address this, we applied domain adaptation, in which the model pretrained using healthy subjects' data was adapted using the target stroke patient's training set to enhance model performance.…”