Multi-Modal Feature Selection with Feature Correlation and Feature Structure Fusion for MCI and AD Classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Multi-modal data integration is crucial for accurate classification of neurological conditions like mild cognitive impairment (MCI) and Alzheimer's disease (AD).
• Existing feature selection methods often overlook the intricate relationships within multi-modal data, potentially limiting classification performance.

Purpose of the Study:

• To develop and validate a novel multi-modal feature selection algorithm, FC2FS, for enhanced classification of MCI and AD.
• To improve upon traditional subject-based feature selection by considering feature node relationships and geometric structures.

Main Methods:

• Proposed FC2FS algorithm fuses feature correlation and feature structure regularization within a multi-task learning framework with low-rank constraints.
• Feature correlation regularization is built upon a similarity matrix of multi-modal features.
• Feature structure regularization leverages manifold learning to capture local geometric structures of feature nodes.

Main Results:

• The FC2FS algorithm achieved high classification accuracies: 91.85% (NC vs. AD), 85.33% (NC vs. late MCI), 78.29% (NC vs. early MCI), and 77.67% (early MCI vs. late MCI).
• The method effectively integrates information from multiple data modalities, outperforming traditional approaches.
• Experimental validation confirmed the efficacy of FC2FS in distinguishing between different cognitive states.

Conclusions:

• FC2FS offers a robust approach to multi-modal feature selection, enhancing the interpretation and performance of MCI and AD classification.
• The algorithm's consideration of feature relationships and geometric structures provides valuable insights for neurological disorder identification.
• This research holds significant reference value for the clinical identification and diagnosis of MCI and AD.