Multi-atlas based representations for Alzheimer's disease diagnosis

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Brain morphometry using magnetic resonance (MR) imaging is crucial for diagnosing Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• Current single-atlas based methods may not capture sufficient anatomical differences for accurate classification.
• Registration errors in single-atlas normalization can limit diagnostic performance.

Purpose of the Study:

• To introduce and evaluate a novel multi-atlas based morphometry method for improved classification of AD and MCI.
• To leverage complementary information from multiple brain atlases to enhance diagnostic accuracy.
• To reduce the impact of registration errors in brain morphometric analysis.

Main Methods:

• Subjects were registered to multiple brain atlases, and adaptive regional features were extracted.
• A correlation and relevance based scheme was used for joint feature selection across atlases.
• Support vector machine (SVM) was employed for final classification.

Main Results:

• The multi-atlas method achieved 91.64% accuracy for Alzheimer's disease (AD) versus normal control (NC) classification.
• Classification accuracy for progressive mild cognitive impairment (p-MCI) versus stable mild cognitive impairment (s-MCI) was 72.41%.
• The proposed multi-atlas approach significantly outperformed traditional single-atlas based methods.

Conclusions:

• Multi-atlas based morphometry offers superior performance compared to single-atlas methods for AD and MCI classification.
• This technique provides complementary anatomical information and mitigates registration errors.
• The findings support the clinical utility of multi-atlas morphometry in neurodegenerative disease diagnosis.