Extracting salient brain patterns for imaging-based classification of neurodegenerative diseases

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Neurodegenerative diseases present complex symptoms difficult for radiologists to quantify.
• Existing automatic morphometric analyses often lack clinically relevant anatomo-functional correlations.

Purpose of the Study:

• To develop a fully automatic image analysis method for objective grading of neurological disorders.
• To reveal discriminative brain patterns associated with neurodegenerative diseases.

Main Methods:

• A novel fusion strategy combining bottom-up (multiscale feature analysis) and top-down (max-margin multiple-kernel optimization) information flows.
• Evaluation using Alzheimer's disease (AD) as a use case on public MRI datasets (OASIS-MIRIAD).

Main Results:

• The method demonstrated improved classification performance, reducing the equal error rate by 6.2% to 13% compared to feature-based morphometry.
• Identified discriminant brain regions strongly correlate with established clinical findings in AD.

Conclusions:

• The proposed method offers an objective approach to diagnosing and grading neurodegenerative diseases.
• This technique provides valuable anatomo-functional insights beyond simple classification.