Deep Learning Approaches for Early Detection of Alzheimer's Disease Using MRI Biomarkers
Abstract
This study presents a novel convolutional neural network (CNN) architecture for the early detection of Alzheimer's disease (AD) using structural MRI data. Our model, designated AD-Net, incorporates multi-scale feature extraction and attention mechanisms to identify subtle morphological changes in the hippocampus and entorhinal cortex. Trained on 2,847 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, AD-Net achieves 94.3% classification accuracy, 93.1% sensitivity, and 95.6% specificity in distinguishing early-stage AD from cognitively normal controls, outperforming existing state-of-the-art methods by 6.2 percentage points. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis confirms that the model focuses on clinically relevant brain regions. These findings suggest that deep learning-based MRI analysis holds significant promise for non-invasive, cost-effective early AD screening.