Advanced Deep Learning Framework for Early Alzheimer’s Disease Classification and Disease Monitoring

Background: Alzheimer's disease (AD) is a progressive brain disorder characterized by distinct stages. Although effective treatment varies based on the degree of severity, there lacks an efficient and noninvasive means of classifying stages of AD. Objectives: This work aimed to develop a clinically effective and generalizable deep-learning model that noninvasively classifies various stages of AD using magnetic resonance imaging (MRI) data. We propose a custom convolutional neural network (CNN) architecture to assess structural variations between AD stages, forming a new direction for Alzheimer's diagnostics. Methods: The dataset used to train the model consisted of MRI scans across four stages: NonDemented, VeryMildDemented, MildDemented, and ModerateDemented. We used extensive data augmentation techniques to address MRI variations and applied dropout and batch normalization layers to improve the model's generalizability. The CNN model consists of four convolutional blocks with pooling and regularization layers and dense layers for multi-class classification. We used early stopping and adaptive learning rate adjustments on a training set of 30,336 images and a validation set of 10,112 images, training the model over ten epochs. Results: The model achieved an overall test accuracy of 96.87% and a near-perfect specificity, precision, recall, and F1 score for each class. The area under the curve (AUC) values for the receiver operating characteristic (ROC) curves were 1.00 for all classes. Conclusion: The model's ideal and interpretable metrics suggest its clinical feasibility for early-stage AD monitoring, potentially improving patient outcomes.