Generation of μ CT images from medical CT scans of carbonate rocks using a diffusion-based model
João Paulo da Ponte Souza, Gabriela Fernandes Matheus, Mateus Basso, Guilherme Furlan Chinelatto, Alexandre Campane Vidal · Applied Computing and Geosciences · 2023
Carbonate rocks are known for their high heterogeneity and textural and compositional complexity. Evaluating their petrophysical properties is thus challenging, especially with limited information. One way to obtain an internal image of such rocks is to scan them with X-ray computed tomography scanners, revealing their internal structures. The problem with this approach is the trade-off between cheap bigger volumes scanned with medical CT, but with lower spatial resolution, and higher spatial resolution with μCT, but higher financial and time costs involved, as well as smaller samples (normally centimeter scales). This work proposes a way to have the volume and cost advantages of the medical CT with the spatial resolution of the μCT using a diffusion probabilistic model to convert medical CT into a μCT. The model was trained with a sample of the Brazilian pre-salt composed by Shrub boundstones and tested with one sample of the same facies and another composed of packstone–grainstones. The results show that the model could replicate both samples well, but worked better when the input sample has the same facies as the training set.