Distance-informed metric learning for Alzheimer's disease staging

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Intermediate biomarkers are crucial for Alzheimer's disease (AD) diagnosis and prognosis.
• Current staging methods may lack precision in differentiating disease progression stages.

Purpose of the Study:

• To develop a novel Alzheimer's disease (AD) staging method.
• To improve classification accuracy for Normal Controls (NC), Mild Cognitive Impairment (MCI), and AD groups.

Main Methods:

• Employed a novel metric learning technique incorporating weak supervisory information.
• Utilized pairwise constraints based on Mini Mental State Examination (MMSE) score disparities.
• Integrated the knowledge that MCI typically falls between NC and AD stages into the classification metric.

Main Results:

• Demonstrated improved classification rates compared to state-of-the-art methods (LMNN, RCA).
• Successfully classified subjects into NC, MCI, and AD groups using the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.
• Validated the effectiveness of incorporating relative MMSE score information via constraints.

Conclusions:

• The proposed metric learning approach enhances AD staging accuracy.
• Weak supervisory information, specifically MMSE score disparities, is valuable for refining classification.
• This method offers a promising tool for more precise AD diagnosis and prognosis.