NSCT与AR-PCNN相结合的医学图像融合研究  

Research on medical image fusion by combining NSCT and AR-PCNN

作　　者：裴高乐 史涛 李世星 PEI Gaole;SHI Tao;LI Shixing(College of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,China;Tianjin Key Laboratory for Control Theory and Applications in Complicated Systems,School of Electrical and Electronic Engineering,Tianjin University of Technology,Tianjin 300384,China;Department of Radiology,Tianjin Beichen Hospital,Tianjin 300400,China)

机构地区：[1]华北理工大学电气工程学院,唐山063210 [2]天津理工大学电气电子工程学院天津市复杂系统控制理论及应用重点实验室,天津300384 [3]天津市北辰医院放射科,天津300400

出　　处：《激光杂志》2022年第10期222-228,共7页Laser Journal

基　　金：国家自然科学基金(No.61203343);河北省自然科学基金(No.F2018209289)。

摘　　要：针对传统医学图像融合技术中所使用的脉冲耦合神经网络(PCNN)模型的内部链接矩阵(L通道链接权值)设置的不合理之处,提出一种用面积比(AR)来设置内部链接矩阵参数配置的方法。采用非下采样轮廓波变换(NSCT)将电子计算机断层扫描(CT)图像和磁共振成像(MRI)图像分解为高频系数和低频系数,以经AR改进的PCNN作为高频融合规则得到高频融合系数,以加权平均作为低频融合规则得到低频融合系数,最后通过NSCT反变换重构出融合图像。在两组实验中,该改进算法相较于Wavelet、Contourlet、NSCT-PCNN算法,平均梯度平均提高了16.42%、4.87%、1.45%,信息熵平均提高了3.46%、1.56%、0.23%,标准差平均提高了4.53%、0.96%、0.26%。实验结果表明,经AR改进后融合的医学图像具有更加丰富的多模态医学影像的互补信息,其品质高于传统的医学图像融合技术所得到的医学融合图像。To address the unreasonable setting of the internal link matrix(L-channel link weights) of the Pulse Coupled Neural Networks(PCNN) model used in traditional medical image fusion techniques, a method is proposed to set the parameter configuration of the internal link matrix using Area Ratio(AR). The Non-subsampled Contourlet Transform(NSCT) was used to decompose the Computed Tomography(CT) and Magnetic Resonance Imaging(MRI) images into high-frequency coefficients and low-frequency coefficients. The high-frequency fusion coefficients were obtained by using the PCNN modified by AR as the high-frequency fusion rule, and the low-frequency fusion coefficients were obtained by using the weighted average as the low-frequency fusion rule. Finally, the fused images were reconstructed by NSCT inverse transform. When compared with the average gradients obtained by Wavelet, Contourlet, and NSCT-PCNN algorithms, it obtained by this improved algorithm was increased by an average of 16.42%, 4.87%, and 1.45% in the two groups, while the information entropies was enhanced by an average of 3.46%, 1.56%, and 0.23%, and the standard deviation was elevated by an average of 4.53%, 0.96%, and 0.26%. The experimental results indicated that the fused medical images by the improvement of AR have richer complementary information of multi-modal medical images, and their quality is higher than the medical fused images obtained by the traditional medical image fusion techniques.

关 键 词：医学图像融合 非下采样轮廓波变换 脉冲耦合神经网络 面积比 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]