Abstract
Alzheimer’s disease (AD) is a progressive neurological condition that leads to dementia. This study presents the SelfAttention Augmented Wasserstein Generative Adversarial Network (SAA-WGAN) for classifying AD stages, utilizing images from the Alzheimer’s disease Neuroimaging Initiative (ADNI). Input images were pre-processed with Fast Guided Median Filter (FGMF) to enhance quality and reduce noise. Data augmentation techniques, including rotations and cropping, addressed class imbalances and improved training diversity. The SAA-WGAN model was validated across varying batch sizes, employing both augmented and non-augmented datasets to assess generalization capabilities. The technique was applied to 1,296 images, achieving a peak accuracy of 99% and demonstrating improved performance over conventional methods in key metrics such as AUC, Precision, and Sensitivity. These results highlight the model’s effectiveness and potential to enhance diagnostic accuracy for Alzheimer’s disease stages.
Keywords: ADNI Dataset, Alzheimer’s Disease, Batch Size, Data Augmentation, Deep Learning, Generative Adversarial Network.