The study presents a parametrically optimized generative adversarial network (POP-GAN) framework for synthetic MRI image generation that addresses the limitations of medical data availability due to privacy, cost, and ethics. POP-GAN is compared with StyleGAN2, mustGAN, and cGAN in generating MRI images with a resolution of 128 × 128 pixels, with parameters such as a learning rate of 1 × 10-4, a dropout rate of 0.3, and a buffer size of 6000. Performance was evaluated by MSE, MAE, PSNR, FID, and clinical realism expert scores. POP-GAN reduced MSE from 6.58 × 10-3 to 4.81 × 10-3, increased PSNR, and reduced FID from 32.91 to 24.36 over the baseline model, with a clinical realism of 4.13 out of 5. cGAN achieved the lowest MAE of 3.50 × 10-3, mustGAN the best resolution fidelity, and StyleGAN2 the highest perceived realism. Parameter optimization and progressive training have improved the quality of synthetic MRIs, while POP-GAN provides a balanced compromise for growing datasets in medical imaging.