Synthetically generated foam CT dataset

This dataset was used in a paper currently under review. It consists of synthetically generated computed tomography (CT) data of foam cylinders.

The foam_ct library [1] was used to generate random cylinders of foam and to discretize them into ground-truth volume data and noise-free projection data. 20 cylinders were generated: 2 were used for testing, 2 were used for validation, and the remaining 16 were used for training. Each cylinder was discretized into a volume of 256 slices of 256 × 256 pixels. Moreover, for each slice, 512 projections of width 384 were generated over a range of 180° in a parallel beam geometry.

The subfolders contain the following data:

• "h5" contains a continuous representation of the foam cylinder
• "vol" contains ground-truth volumes
• "proj" contains noise-free projection data
• "proj_poisson500_gaus50" contains the same projection data as in "proj", but with added Poisson + Gaussian noise with a photon count of 500, and a Gaussian variance of 50
• "proj_blurred_poisson500_gaus50" contains the same projection data as in "proj", but with simulated scintillator blurring and noise. After adding Poisson noise (with photon count 500), the images were blurred with a Gaussian blur kernel with sigma=0.8. After that, Gaussian noise was added (with variance 50).
• "proj_blurred_poisson500_gaus50_ref" and "proj_poisson500_gaus50_ref" contain 1024 noise realizations of the first image of the training set (image 1024), with and without scintillator blurring, respectively. These images can be used as calibration measurements for estimating noise model parameters.

References

[1] Pelt, D. M., Hendriksen, A. A., & Batenburg, K. J. (2022). Foam-like phantoms for comparing tomography algorithms. Journal of Synchrotron Radiation, 29(1), 254-265.