IEEE Transactions on Medical Imaging 2018
LEARN
LEARN: Learned Experts' Assessment-Based Reconstruction Network for Sparse-Data CT
Current or former lab member. Select a name to see all of their projects.
Abstract
Compressive sensing is effective for sparse-data CT, such as few-view CT, tomosynthesis and interior tomography, but choosing the regularization parameters adaptively remains a major open problem. LEARN unfolds a fields-of-experts iterative reconstruction scheme into a fixed number of iterations (50 in the experiments) and trains it from data.
On the Mayo Clinic low-dose challenge data, LEARN outperforms several state-of-the-art methods in artifact reduction, feature preservation and computational speed. Because all regularization terms and parameters of the iterative reconstruction are learned from the training data, the network uses application-oriented knowledge more effectively than competing algorithms.
What’s in the repository
- MATLAB code for data generation, training and testing, built on MatConvNet
- A trained model, a test image and a projection matrix for the demo
Citation
@article{chen2018learn,
title = {LEARN: Learned Experts' Assessment-Based Reconstruction Network for Sparse-Data CT},
author = {Chen, Hu and Zhang, Yi and Chen, Yunjin and Zhang, Junfeng and Zhang, Weihua and Sun, Huaiqiang and Lv, Yang and Liao, Peixi and Zhou, Jiliu and Wang, Ge},
journal = {IEEE Transactions on Medical Imaging},
volume = {37},
number = {6},
pages = {1333--1347},
year = {2018},
doi = {10.1109/TMI.2018.2805692}
}