Transfer Learning for Alzheimer's Disease through Neuroimaging Biomarkers: A Systematic Review.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) poses a significant global healthcare challenge.
• Deep learning (DL) and transfer learning (TL) show promise for AD detection and progression prediction using neuroimaging.
• Early diagnosis is crucial for effective AD management.

Purpose of the Study:

• To systematically review the current state of early AD detection and progression prediction.
• To evaluate the effectiveness of DL models with TL using neuroimaging biomarkers.
• To identify key findings and future research directions.

Main Methods:

• Systematic literature review across five databases.
• Screening of 215 studies published between 2010-2020.
• Inclusion of 13 studies meeting specific criteria.

Main Results:

• Maximum accuracy for AD classification reached 98.20% using 3D convolutional networks and local TL.
• Maximum accuracy for AD prognostic prediction reached 87.78% using pre-trained 3D convolutional network architectures.
• TL significantly enhances the accuracy of early AD diagnostic systems.

Conclusions:

• DL models with TL are effective for early AD detection and progression prediction.
• Future research should focus on improving prognostic prediction accuracy.
• Exploring additional biomarkers (e.g., tau-PET, amyloid-PET) and managing dataset size are critical for advancing AD research.