You are new to the world of radiology and/or AI? No worries, we got the most important questions covered ;-)
What is a CT scan?
A CT scan, or computed tomography, is a computerized x-ray imaging procedure in which a narrow beam of x-rays is aimed at a patient and quickly rotated around the body to capture images from different angles. The collected data can then be assembled to form cross-sectional three-dimensional images.
What are CT scans used for?
Radiologists use CT scans to diagnose and stage cancer, detect internal injuries and disease, and diagnose muscle and bone disorders, among other things.
What are DDPMs?
Denoising diffusion models (DDPMs) are a type of AI that can generate data from pure noise. The process is called diffusion because during training of the model, the data is perturbed by adding various degrees of noise simulating that the data “diffuses” and the model learns to predict the noise that was added to the model. During data sampling the DDPM can then reverse this process by starting from random noise and iteratively removing a little bit of it. This “reverse” diffusion is a stochastic process, thus, different outputs can be generated from the same input noise.
What is conditioning in DDPMs?
During training one can add additional information as input for the DDPM. This allows the model to later generate data given a condition, e.g. a text or segmentation mask. You might have already heard of and maybe even used conditional diffusion models, for example Stable-Diffusion or DALL-E. These are so-called text-to-image AI models that generate an image based on a text prompt.