A Hybrid Model for Alzheimer’s Disease Classification Based on Neural Network Architectures Enhanced by GAN Model

Abstract

Alzheimer’s disease (AD) is a neurodegenerative disorder marked by progressive cognitive decline, making early and accurate diagnosis vital for timely intervention. This study explores the efficacy of combining generative adversarial networks (GANs), convolutional neural networks (CNNs), and vision transformers (ViTs) for AD classification using magnetic resonance imaging (MRI) data. GANs were employed to generate synthetic brain images, addressing data scarcity by augmenting the dataset. CNNs were then used for feature extraction, accelerating model training, and mitigating overfitting. These extracted features were subsequently fed into ViTs, known for their ability to capture spatial dependencies in image data. Experimental results demonstrated that the proposed GAN-CNN-ViT fusion model achieved high accuracy (96%) and robustness, outperforming traditional machine learning (ML) and deep learning approaches. GAN-generated synthetic images enhanced dataset generalization, improving ViT performance in distinguishing AD patients from healthy controls. Comparative analyses validated the superiority of this approach over recent methods in AD classification. This framework underscores the potential of deep learning techniques in advancing neuroimaging-based disease diagnosis. It holds significant promise for early AD detection, ultimately contributing to improved patient outcomes and quality of life through the integration of cutting-edge computer vision and ML methodologies in medical applications.