Unsupervised/Semi-supervised Deep Learning for Low-dose CT Enhancement

• DOI: 10.48550/arxiv.1808.02603
• Publication Date: 2018-01-01
• Authors: Mingrui Geng; Yun Deng; Qian Zhao; Qi Xie; Dong Zeng; Jianhua Ma; Wangmeng Zuo; Deyu Meng
• Keywords: Ground truth

Recently, deep learning(DL) methods have been proposed for the low-dose computed tomography(LdCT) enhancement, and obtain good trade-off between computational efficiency image quality. Most of them need large number pre-collected ground-truth/high-dose sinograms with less noise, train network in a supervised end-to-end manner. This may bring major limitations on these because such low-dose/high-dose training sinogram pairs would affect network's capability sometimes ground-truth are hard to be obtained scale. Since ones relatively easy obtain, it should critical make sources play roles an unsupervised learning To address this issue, we propose DL method LdCT enhancement that incorporates unlabeled directly into training. The effectively considers structure characteristics noise distribution measured sinogram, then learns proper gradient pure Similar labeled ground-truth, information can used sufficient experiments patient data show effectiveness method.