Predicting Amyloid-β Positivity in Alzheimer's Disease: Comprehensive Analysis of Feature Selection and Machine...

Area of Science:

• Neuroscience
• Artificial Intelligence

Background:

• Accurate prediction of brain amyloid-β (Aβ) positivity is essential for identifying individuals at risk of Alzheimer's disease (AD).
• Existing methods require refinement to improve early and precise risk assessment.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting Aβ positivity in diverse cognitive groups.
• To identify key predictors of Aβ positivity using a comprehensive dataset.

Main Methods:

• Utilized data from 1,377 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI), categorized into cognitive normal, subjective memory complaints, early/late mild cognitive impairment, and AD groups.
• Developed and assessed 15 ML models using 17 predictors including demographics, cognitive measures (e.g., ADAS13), and APOE4 status.
• Analyzed model performance across ten sex-based subgroups for comprehensive evaluation.

Main Results:

• ML models demonstrated significant predictive power for Aβ positivity.
• Cognitive assessment measures, particularly ADAS13, were identified as important predictors across multiple models.
• Highest accuracies achieved ranged from 0.739 to 0.903 across the ten subgroups.

Conclusions:

• The developed ML models provide practical risk feature scores for enhanced identification of individuals likely to be Aβ positive.
• These findings support the use of ML in conjunction with cognitive and demographic data for improved Alzheimer's disease risk stratification.