Artificial Brains: Synthetic Brain MRI Generation

Technology(s) Used: GANs (DCGAN, WGAN-GP, UNet-GAN), 3D Deep Learning, Conditional Diffusion Models, PyTorch

• Objective: Addressed limited brain tumor MRI datasets by generating synthetic scans across four modalities (T1C, T1N, T2W, T2-FLAIR) to enhance clinical research capabilities.
• Modeling & Innovation: Implemented and compared multiple generative architectures. Developed a 3D WGAN-GP for volumetric consistency and a conditional diffusion model guided by segmentation masks for high structural fidelity.
• Preprocessing: Standardized 3D MRI volumes through normalization and targeted slice selection focusing on high tumor presence.
• Evaluation: Achieved high-fidelity results with 3D WGAN-GP (FID 37.23, PSNR 26.99) and Diffusion models (SSIM 0.84), ensuring anatomical coherence and realism.