Early functional magnetic resonance imaging activations predict language outcome after stroke

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Accurate prediction of stroke recovery is crucial for personalized rehabilitation.
• Aphasia, a common consequence of stroke, significantly impacts patients' quality of life.
• Individualized therapy requires reliable methods to forecast recovery trajectories.

Purpose of the Study:

• To assess the utility of functional magnetic resonance imaging (fMRI) in predicting language recovery after stroke.
• To explore the potential of machine learning techniques for forecasting individual patient outcomes.
• To determine the optimal timing for neuroimaging assessments to predict recovery.

Main Methods:

• 21 aphasic stroke patients underwent language testing at 2 weeks and 6 months post-stroke.
• fMRI scans from an auditory language comprehension task were analyzed.
• Multivariate machine learning models were employed to predict language performance and improvement.

Main Results:

• fMRI data correctly classified 76% of patients into good and poor language recovery groups.
• Classification accuracy increased to 86% when incorporating age and initial language scores.
• Similar predictive accuracy was achieved for forecasting the degree of language improvement.

Conclusions:

• Machine learning combined with fMRI shows high potential for predicting stroke-related language recovery.
• Assessing brain activation potential in the second week post-stroke yields the best predictive accuracy.
• These findings support tailored, intensive early rehabilitation strategies and potential extension to other recovery systems.