The study presents a new framework called POP-GAN, which optimizes the parameters of artificial neural networks (GANs) for generating synthetic MRI images[1]. The main benefit is the systematic setting of key parameters such as batch size, learning rate, dropout, buffer size and activation functions specifically for MRI data[1]. Research has shown that suboptimal settings lead to limited output diversity and unstable training, which degrades the quality of generated images[1]. POP-GAN achieves higher image quality and more stable training compared to previous methods with lower error rates and increased perception scores[1]. The method combines the progressive growth of GANs with the appropriate setting of normalization layers and activation functions, thereby creating synthetic images with anatomically correct structures and appropriate tissue contrast[1]. This approach provides a repeatable procedure for generating high-quality synthetic MRI images with parameters finely tuned to the specific needs of the MRI data[1].