Stain the Phony Brains: Generating Synthetic Different Medical Images Modalities Using Latent Diffusion Models

Medical image synthesis is gaining popularity using generative models. latent diffusion models have recently shown promise in producing remarkably life-like representations of items. However, their ability to produce medical imagery has not yet been investigated. We investigate the possibility of latent diffusion models for the synthesis of different medical modalities. Initially, we create MRI images of the brains using a pre-trained DALLE-3 model given an input text prompt. Second, we create synthetic images and use 2300 MRI and 1300 CT images to train a stable diffusion model. Through a qualitative study, we assess the synthetic image data by having two impartial specialists MRI and CT modalities to classify randomly selected samples from the created data as authentic (real), fraudulent (phony), or unclear. Findings show that images produced using the diffusion model can convey qualities that are otherwise highly specific to certain medical conditions in MRI and CT modalities. This paper aims to introduce diffusion models rather than state-of-the-art generative models like Generative Adversarial Networks (GANs) and provide impetus for the medical image synthesis research community to investigate their possibilities.