Alzheimer’s Disease Stage Classification Using a 4-Layer CNN and SMOTE for Imbalanced MRI Data

Alzheimer’s disease (AD), the primary cause of dementia in older adults, is a progressive neurodegenerative disorder necessitating timely and precise diagnosis for optimal therapeutic interventions. This study proposes a deep learning (DL) framework to classify AD stages using MRI data obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. To correct class imbalance, the Synthetic Minority Over-Sampling Technique (SMOTE) was applied at the image level, transforming MRI scans into feature vectors for synthetic sample generation before reshaping and normalization. A 2D Convolutional Neural Network (CNN) was designed to categorize MRI images into four classes: Non Demented, Very Mild Demented, Mild Demented, and Moderate Demented. The CNN architecture includes four convolutional layers with ReLU activation, pooling layers, a fully connected layer with dropout regularization (rate of 0.3), and a Softmax at the output layer. Various optimizers and dropout settings were evaluated, with the Adam optimizer yielding the best results. The model obtained a commendable classification accuracy of 99.42%, demonstrating its potential as a robust tool for clinical decision support in the early detection of Alzheimer’s disease.