Multivariate Data Analysis and Machine Learning for Prediction of MCI-to-AD Conversion

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Computational Neuroscience

Background:

• Early diagnosis of Alzheimer's disease (AD) is crucial for effective treatment.
• Mild cognitive impairment (MCI) is a prodromal stage of AD, with 10-15% converting annually.
• Structural magnetic resonance imaging (MRI) processing is a popular approach in AD research.

Purpose of the Study:

• To differentiate MCI converters (MCIc) from MCI non-converters (MCInc) using machine learning (ML).
• To predict conversion rates from baseline MRI data, including cortical thickness (CTH) and hippocampal volume (HCV).
• To evaluate the classification abilities of Support Vector Machine (SVM), Decision Trees (J48), and Naive Bayes (NB) classifiers.

Main Methods:

• Utilized MRI data from 803 participants in the ADNI cohort (AD, MCIc, MCInc, Healthy Controls).
• Developed six ML models based on CTH and HCV using WEKA classifiers (SVM, J48, NB).
• Assessed classification performance in terms of sensitivity and specificity, and prediction accuracy for conversion.

Main Results:

• Achieved up to 71% sensitivity and 56% specificity in classification experiments.
• Demonstrated prediction accuracy of up to 84% for MCI conversion.
• Identified robust and effective results for MCIc identification using multivariate models.

Conclusions:

• Supervised ML techniques show promise in identifying MCI converters from baseline MRI data.
• Multivariate models derived from CTH and HCV show potential as prognostic tools for AD.
• Future research should focus on combining features from the same subjects for improved prognostic accuracy.