3D Few-View CT Image Reconstruction with Deep Learning

Few-view CT imaging is an important approach to reduce the ionizing radiation dose. In this paper, we propose a threedimensional (3D) deep-learning-based method for few-view CT image reconstruction directly from 3D projection data. The large memory requirement is a critical issue for reconstructing an image volume directly from cone-beam projection data. Our proposed method addresses this problem by compressing the 3D input into a latent space in a data-driven fashion, and then image reconstruction can be performed in the compressed latent space with a significantly reduced computational cost. To avoid the overfitting problem, the network is first pre-trained using natural images from the ImageNet, and fine-tuned on a publicly available abdominal CT dataset.

[1]  Michael Unser,et al.  Deep Convolutional Neural Network for Inverse Problems in Imaging , 2016, IEEE Transactions on Image Processing.

[2]  Bruce R. Rosen,et al.  Image reconstruction by domain-transform manifold learning , 2017, Nature.

[3]  Hongming Shan,et al.  Dual Network Architecture for Few-view CT - Trained on ImageNet Data and Transferred for Medical Imaging , 2019, Developments in X-Ray Tomography XII.

[4]  Hu Chen,et al.  LEARN: Learned Experts’ Assessment-Based Reconstruction Network for Sparse-Data CT , 2017, IEEE Transactions on Medical Imaging.

[5]  Jari Korhonen,et al.  Peak signal-to-noise ratio revisited: Is simple beautiful? , 2012, 2012 Fourth International Workshop on Quality of Multimedia Experience.

[6]  Li Fei-Fei,et al.  ImageNet: A large-scale hierarchical image database , 2009, CVPR.

[7]  Eero P. Simoncelli,et al.  Image quality assessment: from error visibility to structural similarity , 2004, IEEE Transactions on Image Processing.

[8]  Ge Wang,et al.  A Perspective on Deep Imaging , 2016, IEEE Access.

[9]  Martín Abadi,et al.  TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems , 2016, ArXiv.

[10]  M. Glas,et al.  Principles of Computerized Tomographic Imaging , 2000 .

[11]  C. Willmott,et al.  Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance , 2005 .

[12]  Hongming Shan,et al.  Deep Encoder-Decoder Adversarial Reconstruction (DEAR) Network for 3D CT from Few-View Data , 2019, Bioengineering.

[13]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.