Prediction of Long-term Cognitive Function After Minor Stroke Using Functional Connectivity.

Area of Science:

• Neuroimaging and computational neuroscience
• Stroke recovery and neurorehabilitation
• Machine learning applications in medicine

Background:

• Cognitive impairment is a common and significant long-term consequence of stroke.
• Predicting long-term cognitive outcomes is crucial for effective patient management and rehabilitation planning.
• Current prediction methods often rely on clinical data, which may not fully capture the complex neural underpinnings of cognitive recovery.

Purpose of the Study:

• To investigate the predictive power of functional MRI (fMRI) connectivity for long-term cognitive function following minor stroke.
• To develop and validate a machine learning model using fMRI data to predict cognitive performance at 36 months poststroke.
• To compare the predictive accuracy of fMRI-based models with traditional clinical data and other functional networks.

Main Methods:

• A cohort of 72 first-ever stroke patients was followed for 36 months.
• A machine learning algorithm (ridge regression) was trained on fMRI-derived functional networks (poststroke cognitive impairment [PSCI] network) at 6 months poststroke to predict cognitive scores at 36 months.
• Prediction accuracy was assessed across four cognitive domains (memory, attention/executive, language, visuospatial) and validated on an independent dataset (n=40).

Main Results:

• The PSCI network-based machine learning model accurately predicted memory (r²=0.67), attention/executive functions (r²=0.73), visuospatial functions (r²=0.55), and language functions (r²=0.48) at 36 months poststroke.
• The predictive performance of the PSCI network model was comparable or superior to models using other functional networks or clinical data.
• Specific cortical regions, including the left superior frontal cortex, were implicated in predicting memory, attention, and visuospatial functions; cortical thickness at 6 months did not correlate with long-term cognitive function.

Conclusions:

• Functional MRI connectivity, specifically the PSCI network, provides a robust basis for predicting long-term cognitive outcomes after stroke.
• Machine learning models utilizing fMRI data can offer valuable insights into predicting individual cognitive trajectories poststroke.
• These findings support the integration of advanced neuroimaging techniques and machine learning in stroke outcome prediction and personalized rehabilitation.