TY  - JOUR
AU  - Yang, Xiaogang
AU  - Kahnt, Maik
AU  - Brückner, Dennis
AU  - Schropp, Andreas
AU  - Fam, Yakub
AU  - Becher, Johannes
AU  - Grunwaldt, Jan-Dierk
AU  - Sheppard, Thomas L.
AU  - Schroer, Christian G.
TI  - Tomographic reconstruction with a generative adversarial network
JO  - Journal of synchrotron radiation
VL  - 27
IS  - 2
SN  - 1600-5775
CY  - [S.l.]
PB  - Wiley-Blackwell
M1  - PUBDB-2020-01013
SP  - 486 - 493
PY  - 2020
AB  - This paper presents a deep learning algorithm for tomographic reconstruction(GANrec). The algorithm uses a generative adversarial network (GAN) to solvethe inverse of the Radon transform directly. It works for independent sinogramswithout additional training steps. The GAN has been developed to fit the inputsinogram with the model sinogram generated from the predicted reconstruction.Good quality reconstructions can be obtained during the minimization of thefitting errors. The reconstruction is a self-training procedure based on thephysics model, instead of on training data. The algorithm showed significantimprovements in the reconstruction accuracy, especially for missing-wedgetomography acquired at less than 180 rotational range. It was also validatedby reconstructing a missing-wedge X-ray ptychographic tomography (PXCT)data set of a macroporous zeolite particle, for which only 51 projections over70 could be collected. The GANrec recovered the 3D pore structure withreasonable quality for further analysis. This reconstruction concept can workuniversally for most of the ill-posed inverse problems if the forward model iswell defined, such as phase retrieval of in-line phase-contrast imaging.
LB  - PUB:(DE-HGF)16
C6  - pmid:32153289
UR  - <Go to ISI:>//WOS:000519725000027
DO  - DOI:10.1107/S1600577520000831
UR  - https://bib-pubdb1.desy.de/record/436634
ER  -