Feature-ranking-based Alzheimer's disease classification from structural MRI.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Machine Learning for Healthcare

Background:

• High-dimensional classification of magnetic resonance imaging (MRI) data is crucial for automatic Alzheimer's disease (AD) diagnosis.
• Existing methods require robust feature selection to accurately identify disease-related patterns.

Purpose of the Study:

• To develop and evaluate a novel feature selection procedure for improved Alzheimer's disease classification using MRI data.
• To determine the optimal number of discriminative features for classification using the Fisher Criterion.

Main Methods:

• Voxel-based morphometry (VBM) to identify gray matter differences between AD patients and healthy controls (HCs).
• T-test based feature ranking combined with Fisher Criterion to select optimal features (voxel clusters).
• Support vector machine (SVM) classification and data fusion techniques applied to selected features.

Main Results:

• The proposed system effectively utilizes voxel clusters as features and ranks them using t-test scores.
• The Fisher Criterion successfully determined the optimal number of top discriminative features.
• The data fusion method enhanced classification performance, achieving competitive results.

Conclusions:

• The developed feature selection and classification system demonstrates high performance in distinguishing Alzheimer's disease from healthy controls using MRI data.
• This approach offers a promising tool for computer-aided diagnosis of Alzheimer's disease.