Synthetic Magnetic Resonance Images with Generative Adversarial Networks
This work addresses data augmentation for medical research, specifically in brain MRI analysis, but is incremental as it experiments with existing GAN architectures and loss functions.
The study tackled the problem of generating synthetic brain MRI images using GANs, finding that hyperparameter tuning, a mini-batch similarity layer, and gradient penalty are crucial for achieving convergence with high realism, though it required significant computation time.
Data augmentation is essential for medical research to increase the size of training datasets and achieve better results. In this work, we experiment three GAN architectures with different loss functions to generate new brain MRIs. The results show the importance of hyperparameter tuning and the use of mini-batch similarity layer in the Discriminator and gradient penalty in the loss function to achieve convergence with high quality and realism. Moreover, huge computation time is needed to generate indistinguishable images from the original dataset.