Conference Paper

Modified minimum cross-entropy algorithm for PET image reconstruction using total variation regularization

Dept. of Biol. Sci. & Med. Eng., Southeast Univ., Nanjing, China
DOI: 10.1109/BIOCAS.2004.1454141 Conference: Biomedical Circuits and Systems, 2004 IEEE International Workshop on
Source: IEEE Xplore


The basic mathematical problem behind PET is an inverse problem. Due to the inherent ill-posedness of this problem, the reconstructed images usually have noise and edge artifacts. How to decrease the noisy effect while preserving the edges is an open problem. In this paper, we propose a new minimum cross-entropy (MXE) image reconstruction method for PET based on the total variation (TV) norm constraint. Here, TV is presented as a regularization function in MXE based reconstruction algorithm. The use of TV is due to the fact that it can effectively reduce the noise in 2D images while preserving edges. The experimental results show that the proposed method is more effective than the common regularized MXE algorithm, especially for noisy projection data.

Download full-text


Available from: Hongqing Zhu, Oct 16, 2015
4 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: The classification step used to assign the appropriate opacity to each voxel is very important in the volume rendering. A new classification algorithm for volume data, which is based on minimum cross entropy, is proposed in this paper. Firstly, the volume data is constructed from a series of sequential two-dimension industrial CT images and the histogram of the volume data is computed. Secondly, the histogram of the volume data is partitioned into different subsections through calculating the accumulated histogram of volume data according to the number of object classes. Thirdly, in each subsection a threshold is computed based on minimum cross entropy. Finally, the opacity of each voxel is assigned by a transfer function, which is split into subsection by these thresholds. Two experimental results from industrial CT images are presented. One is the volume data of vane and the other is volume data of electric drill, from which we can see that the volume rendering results for industrial components are smooth and the image data isn't lost when rotating great angles. In the mean time, this algorithm makes the simulated disassembly of the industrial components performed on the computer successfully.