Visualizing Alzheimer's disease progression in low dimensional manifolds

Area of Science:

• Neuroimaging
• Biomedical Data Visualization
• Machine Learning

Background:

• Tomographic neuroimaging offers high sensitivity for studying brain diseases like Alzheimer's disease (AD).
• Clinical trials (CT) face challenges in utilizing complex 3-D imaging data for primary outcome measures.
• A need exists for intuitive global indices to track AD progression accurately in clinical settings.

Purpose of the Study:

• To develop a novel visualization tool for tracking Alzheimer's disease (AD) progression over time.
• To create an intuitive and explainable index for patients and families.
• To enhance diagnostic accuracy and sensitivity in monitoring AD.

Main Methods:

• Utilized manifold-based nonlinear dimension reduction, specifically Locally Linear Embedding (LLE).
• Applied LLE to longitudinal MRI data from 562 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Reduced high-dimensional MRI features to a two-dimensional space for visualization.

Main Results:

• Developed an LLE map visualizing individual AD progression paths.
• Color-coding indicates the likelihood of an individual being categorized as AD.
• Demonstrated the potential of the LLE map as a tool for tracking AD progression.

Conclusions:

• The proposed LLE-derived visualization tool offers a novel approach to track Alzheimer's disease progression.
• This method provides a more intuitive and potentially more accurate way to monitor disease changes over time.
• The visualization tool can assist clinicians and researchers in understanding and managing AD.