Predicting Post-Stroke Somatosensory Function from Resting-State Functional Connectivity: A Feasibility Study.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Stroke-induced brain damage can affect remote regions, impacting overall function.
• Neuroimaging reveals that brain network disruption, not just lesion location, better predicts stroke impairment.
• Predicting post-stroke functional deficits is crucial for rehabilitation and patient outcomes.

Purpose of the Study:

• To investigate the feasibility of predicting post-stroke somatosensory function using machine learning and brain functional connectivity.
• To assess the predictive power of different functional connectivity models (low-order and high-order) for somatosensory deficits.

Main Methods:

• Functional connectivity was used to model global brain function in 43 chronic stroke survivors.
• Somatosensory impairment was quantified using the Tactile Discrimination Test.
• Linear regression and support vector regression models were applied to predict impairment from functional connectivity data, using both low-order and high-order features.

Main Results:

• A regression model incorporating both low-order and high-order functional connectivity achieved a significant correlation coefficient (r = 0.54, p = 0.0002).
• The study demonstrated the feasibility of using machine learning to predict somatosensory function based on disrupted brain networks.

Conclusions:

• Machine learning models analyzing functional brain networks are feasible for predicting residual somatosensory function after stroke.
• Integrating high- and low-order functional connectivity provides a robust approach for assessing stroke-related deficits.