Fusion of Simple Fully Convolutional Network with 3D Generative Adversarial Network for Alzheimer’s Disease

The progressive neurodegenerative disease known as Alzheimer’s disease (AD) severely reduces cognitive function. Early and precise diagnosis is critical for successful clinical intervention. In this study, we introduce a novel deep learning architecture that blends a 3D Generative Adversarial Network (GAN) with a Simple Fully Convolutional Network (SFCN) to enhance AD classification from structural MRI (Magnetic Resonance Imaging) data. The original dataset, obtained from the OASIS database, consists of 416 MRI volumes, which are supplemented with 10 synthetic 3D brain scans created by a GAN trained on the same distribution. This augmentation solves class imbalance and data scarcity, which are two frequent problems in medical imaging. Our 3D CNN-GAN model outperformed the baseline model trained without GAN augmentation, with a test accuracy of 84.3%, an AUC of 0.92, sensitivity of 91.7%, and specificity of 84.6%, as well as a 22% increase in early-stage AD diagnosis. These results emphasize the clinical utility of leveraging synthetic data to enhance diagnosis based on deep learning, allowing for more accurate and timely identification of AD in real-world medical environments.