CONFERENCE / ICCAIS-2026

Diagnosing Alzheimer’s disease early on is still not easy due to two main issues: the lack of medical imaging data and the serious class imbalance between patients with and without Alzheimer’s disease at each stage of the disease. Typically, deep learning models do not work well with small or skewed datasets, therefore we believe there is a solution to both of these problems using Generative Adversarial Networks (GANs) to create synthetic MRI images as solutions. We trained a GAN using the actual MRI images of brains from patients diagnosed with Alzheimer’s, which allowed us to generate realistic synthetic MRI images (especially for hard-to-find patients in the early stages of Alzheimer’s). By mixing the AI-generated images into the real-world data set of MRI images, we significantly balanced the training dataset for our classifier. After training with the newly balanced dataset, there was significant improvement in the classifier’s performance in terms of accuracy, recall, and F1-scores. In clinical terms, our biggest improvement was early stage Alzheimer’s detection. Our work demonstrates that GAN- generated synthetic medical imaging has the potential to provide a viable method for diagnosing rare diseases that lack sufficient imaging data to support conventional diagnostic methods.