Abstract
<jats:p>This study presents a deep learning-based framework using the ResNet18 architecture for automatic classification of Alzheimer's disease stages from Magnetic Resonance Imaging (MRI) scans. The datasetcomprises Non Demented, Very Mild Demented, Mild Demented, and Moderate Demented—with both original and augmented images to ensure data diversity and balance. The methodology involves several key stages. First, all MRI images were resized to a uniform resolution, normalized, and augmented. The ResNet18 model, a deep residual convolutional neural network pre-trained on ImageNet, was utilized through transfer learning, where the initial layers were frozen to retain low-level feature extraction, and the final fully connected layers were fine-tuned to adapt to the Alzheimer's dataset. The trained model achieved an outstanding test accuracy of 98.18%, indicating strong generalization and discriminative capability. The classification report demonstrates high reliability with F1-scores of 0.99 (Mild Demented), 1.00 (Moderate Demented), 0.98 (Non Demented), and 0.97 (Very Mild Demented).</jats:p>