Material decomposition problem in spectral CT: a transfer deep learning approach

Juan F P J Abascal (1) , N. Ducros (1) , V Pronina , S Bussod , A. Hauptmann (2) , S. Arridge (3) , P. Douek (4) , F. Peyrin (1)

1 Imagerie Tomographique et Radiothérapie
2 CMU - Carnegie Mellon University [Pittsburgh]
3 CMIC - Centre for Medical Image Computing
4 Department of Health and Human Services

Current model-based variational methods used for solving the non-linear material decomposition problem in spectral computed tomog-raphy rely on prior knowledge of the scanner energy response, but this is generally unknown or spatially varying. We propose a two-step deep transfer learning approach that can learn the energy response of the scanner and its variation across the detector pixels. First, we pretrain U-Net on a large data set assuming ideal data, and, second, we fine-tune the pretrained model using few data corresponding to a non-ideal scenario. We assess it on numerical thorax phantoms that comprise soft tissue, bone and kidneys marked with gadolinium, which are built from the kits19 dataset. We find that the proposed method solves the material decomposition problem without prior knowledge of the scanner energy response. We compare our approach to a regularized Gauss-Newton method and obtain a superior image quality.

Mots clés: spectral CT; inverse problem; deep learning; transfer learning

Domaines: Intelligence artificielle [cs.AI]; Imagerie médicale; Physique Médicale [physics.med-ph]

Juan F P J Abascal, N. Ducros, V Pronina, S Bussod, A. Hauptmann, et al.. Material decomposition problem in spectral CT: a transfer deep learning approach. 2020 IEEE 17th International Symposium on Biomedical Imaging Workshops, Apr 2020, Iowa City, United States. ⟨10.1109/ISBIWorkshops50223.2020.9153440⟩. ⟨hal-02587658⟩