MRI-based diagnostic model for Alzheimer's disease using 3D-ResNet.

Area of Science:

• Artificial Intelligence
• Neuroimaging
• Neurology

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder and the primary cause of dementia globally.
• Early and accurate diagnosis of AD is critical for timely intervention, as the disease is currently incurable.
• Advancements in deep learning offer new possibilities for developing sophisticated diagnostic tools.

Purpose of the Study:

• To propose and evaluate a novel deep learning model for classifying cognitive states (AD, MCI, CN) using brain MRI data.
• To enhance the diagnostic capabilities by integrating 3D-ResNet, 3D-CNN, and a spatial attention mechanism (SAM).
• To identify critical brain regions involved in AD classification through interpretable AI methods.

Main Methods:

• Development of a 3D-ResNet architecture incorporating a spatial attention mechanism (SAM).
• Utilized the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset with 800 brain MRI scans.
• Employed data augmentation and cross-validation for robust model training and evaluation on a 7:3 train-test split.

Main Results:

• Achieved 92.33% accuracy in classifying three cognitive states (AD, MCI, CN).
• Demonstrated high accuracy in binary classifications: 97.61% (AD vs. CN), 95.83% (AD vs. MCI), and 93.42% (CN vs. MCI).
• Grad-CAM analysis highlighted the cerebral cortex and hippocampus as key regions for AD classification.

Conclusions:

• The proposed AI model provides a robust and interpretable framework for the early diagnosis of Alzheimer's disease.
• The model's performance surpasses existing state-of-the-art methods, offering significant potential for clinical application.
• Findings support the use of AI in neuroimaging for timely detection and intervention in Alzheimer's disease.