Latent feature representation learning for Alzheimer's disease classification.

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Early detection and treatment of Alzheimer's Disease (AD) are critical for patient outcomes.
• Multi-modality imaging data has shown promise in advancing automatic AD diagnosis.

Purpose of the Study:

• To propose a novel method for Alzheimer's Disease classification using latent feature fusion.
• To effectively leverage information from multi-modality imaging data for improved diagnostic accuracy.

Main Methods:

• A latent feature fusion approach is proposed.
• A specific projection matrix is learned for each imaging modality.
• Binary label matrices and local geometry constraints are introduced to project features into a low-dimensional space.
• Latent feature representations from different modalities are fused for classification.

Main Results:

• The proposed method demonstrates effectiveness in classifying Alzheimer's Disease.
• Experimental validation was conducted using the Alzheimer's Disease Neuroimaging Initiative database.

Conclusions:

• Latent feature fusion is a promising technique for Alzheimer's Disease diagnosis using multi-modality imaging.
• The developed method offers an effective approach for automatic AD classification.