Predicting poststroke dyskinesia with resting-state functional connectivity in the motor network

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Evaluating motor function is crucial for stroke rehabilitation.
• Neuroimaging and machine learning can assess patient functional status.
• Predicting stroke-induced dyskinesia severity requires further investigation into individual brain function.

Purpose of the Study:

• Investigate motor network reorganization in stroke patients.
• Develop a machine learning method to predict motor dysfunction.
• Correlate brain function with dyskinesia severity.

Main Methods:

• Used near-infrared spectroscopy (NIRS) to measure motor cortex hemodynamic signals in resting state.
• Analyzed motor network characteristics using graph theory.
• Constructed support vector machine (SVM) models using small-world properties as features.

Main Results:

• Significant differences in small-world properties (clustering coefficient, local efficiency, transitivity, global efficiency) were observed between healthy subjects and stroke patients with varying dyskinesia levels.
• These properties linearly correlated with Fugl-Meyer Assessment scores.
• SVM models achieved 85.7% accuracy in classifying subject groups.

Conclusions:

• NIRS, resting-state functional connectivity, and SVM provide an effective approach for individual-level assessment of post-stroke dyskinesia.
• This method aids in understanding motor network reorganization after stroke.
• The findings support the use of NIRS-based machine learning for clinical assessment.