Synthetic CT noise emulation in the raw data domain

Reducing radiation dose associated with X-ray computed tomography (CT) scans is of key clinical interest. As such, there is active research in methods of generating diagnostically meaningful images using lower dose data, including data pre-processing and advanced reconstruction algorithms. When developing such procedures, it would be ideal from a benchmarking perspective to have multiple acquisitions at varying dose levels or protocols. While such data can be reasonably acquired for phantoms, comparable clinical data is less available. Therefore, it would be very beneficial to develop a method of synthetically adding noise to existing scans in such a way that the resulting data sets yield reasonably accurate noise realizations. We present such a method in this paper that synthetically adds noise directly in the raw data domain (i.e., before the negative log and associated preprocessing).