Ascending Complexity Task GAN and 3D Dense Convolutional Networks for Binary Classification of Alzheimer’s Disease

Medical imaging technologies have made significant advancements, enabling Magnetic Resonance Imaging (MRI) to capture even the subtle structural and functional changes in the brain. This has greatly improved diagnosing of Alzheimer’s disease (AD). However, multimodal images, which are necessary for accurate diagnosis, are often incomplete due to the high cost or unavailability of PET images. To address this issue, we propose a deep learning framework that combines Generative Adversarial Networks (GAN) and Deep 3D Convolutional Neural Networks (3D-CNN) to classify three classes of ADNI subjects, namely AD, CN, and MCI. The framework consists of two main components: ACT-GAN and 3D-Deep CNN. ACT-GAN is used to extract features from each sagittal, axial, and coronal plane and generate fake images in four levels. The discriminator then compares the generator’s output with real images, which are inputted into the 3D-Deep CNN for classification. Simulation results demonstrate that our proposed deep learning architecture produces 93.18% classification accuracy, 70% Sensitivity, 78% Specificity for AD versus CN, 80.27% classification accuracy, 80% Sensitivity, 79% Specificity for MCI versus AD, and 86.24% classification accuracy, 82% Sensitivity, 78% Specificity for MCI versus CN which is highest when compared with three state-of-the-art techniques.