Emphasis Learning, Features Repetition in Width Instead of Length to Improve Classification Performance: Case...

Area of Science:

• Neuroimaging and Computational Neuroscience
• Medical Informatics and Artificial Intelligence

Background:

• Computer-aided diagnosis for Alzheimer's disease (AD) has advanced significantly, often integrating features from MRI, PET, and CSF.
• Current high-performance classification models (>90% accuracy) face challenges in further performance improvement.

Purpose of the Study:

• To propose a novel methodology for enhancing Alzheimer's disease (AD) diagnosis classification accuracy.
• To introduce a unique approach based on feature space replication rather than traditional sample space methods.

Main Methods:

• Feature extraction using VBM-SPM and embedding via concatenation to create subject feature vectors.
• Dimensionality reduction using Principal Component Analysis (PCA) to form a compact feature space.
• A novel iterative replication process on selected feature components, evaluated by Support Vector Machine (SVM) classifiers until performance convergence.

Main Results:

• Achieved state-of-the-art classification accuracies: 98.81% for AD vs. normal control (NC), 81.61% for MCI vs. NC, and 81.40% for AD vs. MCI.
• The proposed feature replication method demonstrated superior performance compared to existing literature for AD diagnosis.

Conclusions:

• The novel feature space replication methodology offers a significant advancement in computer-aided diagnosis for Alzheimer's disease.
• This approach effectively identifies key features and optimizes classification models for distinguishing between normal controls, MCI, and AD patients.