View-centralized multi-atlas classification for Alzheimer's disease diagnosis

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Machine Learning for Healthcare

Background:

• Multi-atlas methods improve Alzheimer's disease (AD) and mild cognitive impairment (MCI) classification over single-atlas approaches.
• Existing multi-atlas methods often average or concatenate features, potentially losing crucial anatomical diagnostic information.
• There is a need for methods that better leverage diverse information from multiple atlases.

Purpose of the Study:

• To propose a novel view-centralized multi-atlas classification method for improved AD and MCI diagnosis.
• To exploit and select discriminative features from multiple atlas spaces more effectively.
• To enhance the accuracy of classifying Alzheimer's disease and its prodromal stages.

Main Methods:

• Brain images were registered to multiple atlases to extract feature representations.
• A view-centralized multi-atlas feature selection method was employed, using inter-atlas guidance.
• Support Vector Machine (SVM) classifiers were trained on selected features, and results were combined via ensemble learning.

Main Results:

• The method achieved 92.51% accuracy for Alzheimer's disease versus normal control classification.
• It reached 78.88% accuracy for progressive MCI versus stable MCI classification.
• The proposed approach demonstrated superior performance compared to previous multi-atlas classification techniques.

Conclusions:

• The novel view-centralized multi-atlas method effectively utilizes anatomical information from multiple atlases.
• This approach significantly improves the classification accuracy for Alzheimer's disease and mild cognitive impairment.
• The findings suggest a promising direction for neurodegenerative disease diagnosis using advanced machine learning.