MRI-based Alzheimer's disease classification using Vision Transformer and time-series transformer: A step-by-step...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's Disease (AD) poses a significant global health challenge.
• Accurate and early diagnosis of AD is crucial for effective management.
• Structural brain Magnetic Resonance Imaging (MRI) offers valuable insights into neurodegeneration.

Purpose of the Study:

• To develop and validate a reproducible pipeline for classifying Alzheimer's Disease using structural brain MRI.
• To leverage a joint transformer architecture integrating Vision Transformer (ViT) and Time-Series Transformer (TST) models.
• To assess the framework's performance in binary (AD vs. Normal Control) and multiclass (AD, Mild Cognitive Impairment, Normal Control) classification tasks.

Main Methods:

• A novel pipeline utilizing a joint transformer architecture combining ViT and TST models was developed.
• Pre-trained ViT was employed for feature extraction from 2D MRI slices.
• Sequential modeling with a transformer-based classifier captured inter-slice dependencies for classification.

Main Results:

• The proposed framework demonstrated effective classification of Alzheimer's Disease.
• Performance was evaluated on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Successful classification was achieved across axial, sagittal, and coronal MRI planes.

Conclusions:

• The developed reproducible pipeline offers a promising approach for Alzheimer's Disease classification from structural brain MRI.
• The joint transformer architecture effectively captures spatial and sequential information within MRI volumes.
• This method holds potential for improving early diagnosis and monitoring of Alzheimer's Disease.