Improving Alzheimer's stage categorization with Convolutional Neural Network using transfer learning and different...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's Disease (AD) is a progressive neurodegenerative disorder impacting memory and cognition.
• Multi-modal imaging (structural MRI, DTI) aids in identifying brain biomarkers for AD classification.
• Convolutional Neural Networks (CNNs) show promise in enhancing image-based classification tasks.

Purpose of the Study:

• To develop a transfer learning scheme using CNNs for automated Alzheimer's Disease (AD) classification.
• To evaluate the efficacy of transfer learning on limited brain scan data and small regions of interest (ROIs).
• To assess cross-modal, cross-domain, and hybrid transfer learning strategies for AD diagnosis.

Main Methods:

• A LeNet-like CNN architecture was employed for AD stage classification.
• Transfer learning was implemented using cross-modal (sMRI and DTI) and cross-domain (MNIST) approaches.
• The methodology focused on small ROIs, specifically a few slices of the hippocampal region.

Main Results:

• The proposed method demonstrated strong performance on small datasets and with limited brain slices.
• Classification accuracy improved by over 5 points for challenging tasks like AD/MCI and MCI/NC.
• Good accuracy scores were achieved using a shallow convolutional network and cross-modal transfer learning.

Conclusions:

• The developed method is effective for shallow CNNs applied to low-resolution MRI and DTI scans.
• Significant results were obtained even when training on small datasets, common in medical image analysis.
• This approach is suitable for Alzheimer's Disease classification in resource-limited scenarios.