Predicting cognitive decline in a low-dimensional representation of brain morphology

Area of Science:

• Neuroscience
• Medical Imaging Analysis
• Computational Biology

Background:

• Early identification of neurodegeneration in Alzheimer's disease (AD) is crucial for intervention.
• Cortical thickness from MRI is a sensitive marker, but aging and variability pose challenges.
• Analyzing neurodegenerative trajectories requires advanced analytical methods.

Purpose of the Study:

• To develop a novel method for visualizing and analyzing neurodegenerative trajectories using cortical thickness data.
• To differentiate between healthy aging, mild cognitive impairment (MCI), and AD based on these trajectories.
• To assess the potential of this method for early AD detection.

Main Methods:

• Utilized Uniform Manifold Approximation and Projection (UMAP), a nonlinear dimension reduction technique.
• Trained embeddings on cortical thickness data from 6237 healthy individuals and 233 MCI/AD patients (ADNI database).
• Projected longitudinal MRI data into a 2D manifold and analyzed trajectories using clustering and statistical comparisons.

Main Results:

• The primary axis of the UMAP embedding correlated positively with participant age.
• Distinct average trajectories were observed between clusters for MCI and AD subjects.
• Specific clusters and trajectories showed a higher prevalence of AD subjects, achieving an AUC of 0.80 for differentiation.

Conclusions:

• The UMAP-based 2D representation effectively captures neurodegenerative progression.
• This approach demonstrates significant potential for distinguishing MCI and AD patients based on their brain structure trajectories.
• The findings support the use of advanced dimensionality reduction for early Alzheimer's disease diagnosis.