A Classification Algorithm by Combination of Feature Decomposition and Kernel Discriminant Analysis (KDA) for...

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Magnetic resonance (MR) imaging detects brain changes in Alzheimer's disease (AD) and mild cognitive impairment (MCI).
• Distinguishing between AD, MCI, and normal aging is challenging due to overlapping brain image alterations.
• Existing machine learning methods struggle with feature extraction for accurate classification.

Purpose of the Study:

• To develop an automatic classification technique for differentiating between progressive MCI (pMCI) vs. normal control (NC), AD vs. NC, and pMCI vs. stable MCI (sMCI).
• To improve the accuracy of computer-aided diagnosis for neurodegenerative diseases using MR image analysis.

Main Methods:

• Proposed an automatic classification technique combining feature decomposition via dictionary learning and kernel discriminant analysis (KDA).
• Feature decomposition separates class-specific from non-class-specific components.
• KDA maps nonlinearly separable features to a linearly separable space.

Main Results:

• The method was evaluated on T1-weighted MR images from 830 subjects (ADNI dataset).
• Achieved classification accuracies: 90.41% (AD vs. NC), 84.29% (pMCI vs. NC), and 65.94% (pMCI vs. sMCI).

Conclusions:

• The proposed feature decomposition and KDA technique shows promising performance for classifying AD and MCI stages.
• This automated approach can aid in the early and accurate diagnosis of neurodegenerative conditions.