Machine learning-based prediction of mild cognitive impairment among individuals with normal cognitive function

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Previous research primarily identified risk factors for mild cognitive impairment (MCI) and dementia.
• This study focuses on identifying early risk factors for MCI in cognitively normal individuals to enable preventative strategies.

Purpose of the Study:

• To develop a predictive model for the future risk of MCI in a cognitive normal population.
• To identify key clinical and imaging variables associated with MCI development.

Main Methods:

• A longitudinal retrospective study utilizing brain MRI scans, clinical data, and cognitive assessments over 3+ years.
• Application of multiple machine learning algorithms including random forest, support vector machine, logistic regression, eXtreme Gradient Boosting (XGB), and naïve Bayes.
• Development of prediction models using combinations of clinical and imaging variables.

Main Results:

• The eXtreme Gradient Boosting (XGB) model demonstrated superior classification performance.
• Prediction accuracy reached 94.32% when combining clinical and imaging variables.
• Key predictors identified were white matter hyperintensity (WMH) degree (especially in the frontal lobe) and systolic blood pressure (SBP) control.

Conclusions:

• The XGB model integrating clinical and imaging data offers a promising approach for MCI risk prediction.
• Frontal lobe white matter hyperintensity and systolic blood pressure control are critical factors in predicting MCI development.