Multimodal Data Analysis of Alzheimer's Disease Based on Clustering Evolutionary Random Forest

Area of Science:

• Neuroscience
• Genetics
• Data Science

Background:

• Alzheimer's disease (AD) presents a significant global health challenge.
• Technological advancements yield high-dimensional multimodal data, including functional magnetic resonance imaging (fMRI) and single nucleotide polymorphism (SNP).
• Understanding complex associations within this heterogeneous data is crucial for AD diagnosis and prevention.

Purpose of the Study:

• To develop a comprehensive diagnostic framework for Alzheimer's disease using multimodal data.
• To identify significant associations between brain regions and genes relevant to AD.
• To improve the generalization performance and diversity of machine learning models for AD classification.

Main Methods:

• Correlation analysis to detect relationships between brain regions (fMRI) and genes (SNP), creating "brain region-gene pairs" as multimodal features.
• Development and application of Cluster Evolutionary Random Forest (CERF), a novel method enhancing random forest generalization through hierarchical clustering of decision trees.
• Integration of feature construction, selection, and sample classification within the CERF framework to optimize diagnostic performance.

Main Results:

• The proposed framework effectively identified Alzheimer's disease patients using fMRI and SNP data from the ADNI database.
• Significant brain regions and genes associated with Alzheimer's disease were discovered.
• The CERF method demonstrated improved generalization by clustering similar decision trees and retaining high-performing ones.

Conclusions:

• The developed diagnostic framework offers a promising approach for Alzheimer's disease identification.
• The identified brain region-gene pairs provide valuable insights into AD pathophysiology.
• This multimodal data analysis strategy contributes to the clinical treatment and prevention of Alzheimer's disease.