Evaluation and Prediction of Early Alzheimer's Disease Using a Machine Learning-based Optimized Combination-Feature...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Alzheimer's Disease (AD) presents complex pathological changes, making evaluation challenging with traditional methods.
• Machine learning (ML) offers novel approaches to overcome limitations in AD assessment.
• Gray Matter Volume (GMV) and Quantitative Susceptibility Mapping (QSM) are advanced imaging techniques with potential for AD analysis.

Purpose of the Study:

• To investigate ML-based classification and prediction using an Optimized Combination-Feature (OCF) set.
• To utilize GMV and QSM data for differentiating between Cognitive Normal (CN), Amnestic Mild Cognitive Impairment (aMCI), and AD subjects.
• To assess the efficacy of ML models in predicting the aMCI stage.

Main Methods:

• Utilized Support Vector Machine (SVM) with OCF sets of GMV and QSM data for group classification (CN, aMCI, AD).
• Employed regression-based ML models, including Gaussian process regression, for predicting aMCI stages.
• Defined Regions-of-Interest (ROIs) in iron-rich and amyloid-prone brain areas.

Main Results:

• Achieved high classification accuracy between groups: CN vs. aMCI (AUC=0.94), aMCI vs. AD (AUC=0.93), and CN vs. AD (AUC=0.99).
• The Gaussian process regression model demonstrated prediction accuracy for aMCI closely matching clinical data (RMSE=0.371 vs. 0.319).
• Optimal classification utilized specific combinations of GMV and QSM data from hippocampus, entorhinal cortex, amygdala, pulvinar, and posterior cingulate cortex.

Conclusions:

• The OCF-based ML approach using GMV and QSM shows significant effectiveness in classifying subject groups.
• This method demonstrates strong potential for predicting the aMCI stage, correlating well with clinical assessments.
• The developed ML approach can serve as a valuable tool for personalized analysis and diagnostic aid in Alzheimer's Disease.