文章摘要

文章信息/Info

Title:: Bidirectional multi-resolution demons algorithm ?based on local joint entropy gradient

Author(s):: WANG Chang; REN Qiongqiong; QIN Xin; LIU Yan; ZHANG Wenchao; YU Yi; School of Biomedical Engineering, Xinxiang Medical University, Xinxiang 453003,Henan Province，China

Keywords:: Bidirectional; Multi-resolution registration; Local joint entropy ; Active demons; Diffeomorphic demons

摘要:: 针对active demons算法易受到参数设置的影响，无法有效解决大形变场的配准问题，本研究提出了基于局部联合熵梯度的双向多分辨率demons算法。利用在配准过程中两幅图像的互信息不断增加，局部联合熵增大的规律，本研究引入两幅图像局部联合熵参数，将图像局部联合熵的梯度附加到demons驱动力中，实现了基于局部联合熵梯度的双向多分辨率demons算法。利用自然图像、MRI图像和CT图像测试本算法的优越性，〖JP2〗与active demons, diffeomorphic demons算法进行对比分析，采用均方误差、归一化互相关系数和结构相似度对配准结果进行定量评价。本算法归一化互相关系数和结构相似度最高，均方误差最小。通过分析权重系数的影响和设置合适的参数，本算法可应用于大形变的医学图像配准，具有一定的临床应用价值。

Abstract:: Active demons cannot effectively solve registration problem with large deformation, and the result were influenced by parameter setting, bidirectional multi-resolution demons based on local joint entropy gradient was proposed in this paper. In the process of registration, mutual information will increase, and local joint entropy will enlarge. Based on this rule, local joint entropy parameter of two image was introduced. The gradient of local joint entropy was added to demons driving force, bidirectional multi-resolution demons based on local joint entropy gradient was realized. The superiority of this algorithm was verified by natural image, MRI and CT image comparing with active demons, diffeomorphic demons. Mean square error, normalized cross correlation and structural similarity were used as statistical analysis indicators in the process of evaluation of registration result. Normalized cross correlation and structural similarity of this algorithm is the highest, mean square error is the lowest by this method. By analyzing influence of weight coefficient on registration result and setting appropriate parameters, this method can be applied to medical image registration with large deformation. The research result is helpful for clinical application.