Discriminative patterns of white matter changes in Alzheimer's

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Structural brain connectivity changes are crucial in Alzheimer's disease (AD), but advanced analytical methods are underexplored.
• Understanding these alterations is key for developing effective diagnostic tools.

Purpose of the Study:

• To develop and apply an efficient structural connectome-based convolutional neural network (CNN) for classifying AD.
• To identify and visualize discriminative white matter connectivity changes in AD using CNN explanations.

Main Methods:

• A CNN architecture was designed to classify normal control (NC) and AD subjects based on weighted structural connectomes.
• Gradient-based localization techniques were employed to analyze CNN decisions and pinpoint specific connectivity alterations.

Main Results:

• The CNN successfully classified AD subjects from NC subjects using structural connectome data.
• Discriminative white matter connectivity changes were identified, predominantly involving temporal/subcortical regions.
• A distinct pattern of altered connectivity was observed within and between temporal/subcortical, frontal, and parietal regions in both hemispheres.

Conclusions:

• The developed CNN approach effectively identifies discriminative structural connectivity differences in AD.
• This methodology offers a novel way to detect AD biomarkers and understand brain changes.
• The findings can enhance clinician confidence in automated diagnostic systems for AD.