Tensor-Based Multi-Modality Feature Selection and Regression for Alzheimer's Disease Diagnosis

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Machine Learning in Medicine

Background:

• Accurate assessment of Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI) using brain imaging is challenging.
• Combining multi-modality imaging techniques enhances the reflection of pathological characteristics for improved AD and MCI diagnosis.

Purpose of the Study:

• To propose a novel tensor-based multi-modality feature selection and regression method for AD and MCI diagnosis.
• To identify diagnostic biomarkers for AD and MCI by leveraging high-level correlations in multi-modality data.

Main Methods:

• Utilized a tensor structure to exploit correlations within multi-modality data (VBM-MRI, FDG-PET, AV45-PET).
• Investigated tensor-level sparsity in a multilinear regression model for feature selection and diagnosis.
• Applied the method to ADNI data, incorporating clinical parameters of disease severity and cognitive scores.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art approaches for disease diagnosis.
• Successfully identified disease-specific brain regions and modality-related differences.
• Validated practical advantages using real-world ADNI data.

Conclusions:

• The novel tensor-based multi-modality approach offers improved accuracy in diagnosing AD and MCI.
• This method is effective for biomarker identification and understanding modality-specific contributions to disease pathology.
• The publicly available code facilitates further research in neurodegenerative disease diagnostics.