Super-Resolution reconstruction of Bone Micro-Structure Micro-CT Image based on Auto-Encoder Structure
Xuyang Xie, Yu Wang, Shibo Li, Long Lei, Ying Hu, Jianwei Zhang · 2019
Computed Tomography (CT) is widely used for screening, diagnostic and image-guided therapy for clinical and research purposes. The current Finite Element Analysis for Research Microstructure based on Micro-CT scan data has been considered the gold standard for non-invasive studies of bone microstructure. In medical diagnosis and treatment, the resolution of clinical medical images is low, and it is impossible to directly observe the fine structure of bone through Micro-CT. However, there are many problems in high resolution image, such as long scan time, high radiation dose and complicated processing steps. Therefore, the research goal of this paper is to propose a new network structure which can reconstruct high-resolution bone microstructure images with fine structure while reducing X-ray radiation. The network model proposed in this paper combines the auto-encoder structure and adopts a staged up-sampling strategy to accurately reconstruct high-resolution images from low-resolution images.Moreover, in order to be more realistic, this paper constructs a real training data set according to strict operation. In the data set of this paper, a variety of existing super-resolution reconstruction methods are used for comparison test. The results show that the network model proposed in this paper achieves better reconstruction results. We found in the experiment that the noise of the sample itself has a certain influence on the evaluation of the reconstruction effect and the impact of noise reduction is also discussed on super-resolution reconstruction.