Attention-Guided Generative Adversarial Networks for Enhancing MRI-based Alzheimer's Disease Diagnosis

o date, Alzheimer's disease (AD) remains a major neurodegenerative disorder without a cure, and early and accurate diagnosis is critical for patient management. To this end, we propose an Attention Guided Generative Adversarial Network (AG-GAN) framework for improving MRI-based Alzheimer's disease diagnosis in this study. The dual-stream attention model is proposed, in which the deep learning process is refined by a dual-stream attention mechanism to extract refined MRI features and enhance its capability of discriminability to labels. The AG-GAN framework provides high-quality synthetic MRI scans that help with data augmentation and mitigate the class imbalance problem in AD datasets. Moreover, the attention module is designed to learn key pathological features using robust feature learning. Finally, we test the model on AD datasets and benchmark it against baseline CNN classifiers and standard GAN-based classifiers. The results of the proposed experiment show improved classification accuracy, sensitivity, and specificity of this approach for early AD detection. This work contributes to developing deep learning techniques in medical imaging applications to increase model interpretability and diagnostic precision.