Nonlinear dimensionality reduction combining MR imaging with non-imaging information.

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Computational Neuroscience

Background:

• Inter-subject brain variation analysis is crucial for understanding neurological disorders.
• Current biomarker extraction methods often rely on predefined or segmented features.
• Integrating diverse data types can improve diagnostic accuracy.

Purpose of the Study:

• To propose a data-driven framework for extracting biomarkers from low-dimensional brain manifolds.
• To incorporate subject meta-information into manifold learning for enhanced classification.
• To validate the framework using Alzheimer's Disease (AD) and mild cognitive impairment (MCI) classification.

Main Methods:

• Developed a manifold learning framework extending Laplacian Eigenmaps.
• Integrated non-imaging data (ApoE genotype, CSF Aβ(1-42)) and imaging-derived biomarkers (hippocampal volume) into the learning process.
• Tested classification performance on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database.

Main Results:

• The proposed method achieved favorable classification results for AD, MCI, and healthy controls.
• Data-driven biomarkers were extracted, offering a unified representation of imaging and non-imaging data.
• Performance compared favorably to established neuroimaging methods.

Conclusions:

• The framework provides a unified, data-driven approach to biomarker discovery.
• Integrating subject meta-information significantly enhances manifold learning for clinical applications.
• This method shows promise for improved diagnosis and understanding of neurodegenerative diseases.