PVTAD: ALZHEIMER'S DISEASE DIAGNOSIS USING PYRAMID VISION TRANSFORMER APPLIED TO WHITE MATTER OF T1-WEIGHTED...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) is a neurodegenerative disorder requiring early diagnosis for effective intervention.
• Disrupted neural connections in AD offer potential biomarkers, but current machine learning models struggle to capture complex patterns.
• Existing methods using Convolutional Neural Networks (CNN) and Vision Transformers (ViT) may not fully exploit local and global brain features.

Purpose of the Study:

• To introduce PVTAD, a novel approach for classifying Alzheimer's disease (AD) and cognitively normal (CN) individuals.
• To leverage pretrained Pyramid Vision Transformer (PVT) and white matter (WM) features from structural MRI (sMRI) for improved AD detection.
• To combine the strengths of CNN and ViT for comprehensive feature extraction from WM in sMRI scans.

Main Methods:

• Utilized T1-weighted MPRAGE sMRI scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Developed the PVTAD model, integrating PVT with WM coronal middle slices to capture local and global AD-indicative patterns.
• Compared PVTAD against single and parallel CNN and standard ViT models for AD vs. CN classification.

Main Results:

• PVTAD achieved a high classification accuracy of 97.7% for AD vs. CN.
• The model demonstrated a strong F1-score of 97.6%, indicating robust performance.
• PVTAD significantly outperformed traditional CNN and ViT models in classifying AD using sMRI data.

Conclusions:

• The PVTAD approach shows significant promise for accurate and early Alzheimer's disease diagnosis using sMRI.
• Integrating PVT with WM analysis offers a superior method for capturing AD-related neuroimaging biomarkers.
• The developed model provides a valuable tool for advancing AD research and clinical applications.