Design of a CNN–Swin Transformer Model for Alzheimer’s Disease Prediction Using MRI Images

Abstract

Alzheimer’s Disease (AD) is a progressive neurological condition that deteriorates memory, cognition, and behavior, especially in older adults. Timely identification is essential to enhance patient outcomes and inform therapy choices. This research introduces an extensive deep learning model for multiclass Alzheimer’s disease stage classification using structural MRI images from the publicly accessible OASIS dataset. The procedure begins with thorough preprocessing, including skull stripping, axial slice extraction, and intensity normalization to guarantee uniform input quality. Deep Convolution Generative Adversarial Network (DCGAN) is used to produce authentic synthetic MRI slices, therefore addressing data imbalance and enhancing class representation and training stability. The EffSwin-XNet model introduces a novel hybrid deep learning framework that strategically fuses EfficientNet-B0 and the Swin Transformer, enabling both local and global feature extraction from MRI brain images. This represents a significant advancement over conventional convolutional neural networks and a feature fusion attention mechanism that adaptively emphasizes discriminative features. Grad-CAM is used for explainability to view the brain areas influencing each categorization judgment, hence increasing therapeutic confidence. The model attains a classification accuracy of 95.3%, surpassing traditional CNN and hybrid benchmarks. This study presents an enhanced, interpretable, and efficient method for stage-wise categorization of Alzheimer’s Disease, demonstrating significant potential for use in clinical decision-support systems for early classification and intervention.

Topics & Concepts