基于PCA与二代Curvelet变换的多模态医学图像融合方法研究

The multimodal medical image fusion is the current research hot spot and difficulty in the field of medical image processing. The second generation Curvelet transform that effectively overcomes the shortcoming of the traditional two-dimensional wavelet transform in the direction and sparsity reflects the latest developments of multiscale geometric analysis in recent years. Based on our preliminary studies, this project will focus on the Curvelet transform and its applications in the fusion of structural and functional medical images for the following research. Firstly, in view of the shift-variant defect of the Curvelet transform and the characteristics of the medical images to be fused, a novel shift-invariant multimodal medical image fusion method is presented which is based on the PCA (Principal component analysis) and Curvelet transform. Secondly, in order to guarantee the robustness and adaptability of the fusion process, different fusion strategies are developed for the decomposed coefficients of low and high frequency bands according to their different physical properties. In the project, a regional similarity based method is proposed to fuse the low frequency coefficients, and a neighborhood entropy based PCNN (Pulse-coupled neural network) method is put forward to fuse the high frequency coefficients. Finally, based on the characteristics of the human visual system, we will study a new quality assessment of the fusion. The research of this project will provide new ideas and new techniques for fusion of structural and functional medical images and its result can effectively improve the quality and performance of medical image fusion. Therefore, the study of this project on multimodal medical image fusion will play a key role in clinical diagnosis, medical research, and so on.

多模态医学图像融合是当前医学图像处理领域中的研究热点和难点。第二代Curvelet变换有效克服了传统二维小波变换在方向性、稀疏性等方面的不足，反映了近年来图像多尺度几何分析的最新进展。本项目在前期研究基础上，针对Curvelet变换及其在结构和功能医学图像融合中的应用进行如下研究：针对Curvelet变换的移变性缺陷并结合融合医学图像的特性提出一种基于PCA+Curvelet变换且具有"移不变性"的多模态医学图像融合新方法；为保证融合过程的鲁棒性和自适应性，针对分解系数的低、高频不同物理特性，提出一种基于区域相似度的低频系数融合方法和一种基于邻域熵的PCNN高频系数融合方法；最后，基于人类视觉系统特性，研究新的融合质量评价方案。本项目的研究将为结构和功能医学图像融合提供新的研究思路和方法，其结果将进一步提高医学图像融合的质量和性能，对于多模态医学图像融合在临床诊断和医学研究上具有重要价值。

多模态医学图像融合是当前医学图像处理领域中的研究热点和难点。本项目通过对现有多尺度变换方法在图像融合中的系统研究和问题分析，提出了新的图像融合方法，并将其成功地应用于医学图像的融合。项目的主要创新工作包含：①基于第二代Curvelet变换的优良性能，提出一种基于快速离散Curvelet变换(FDCT)的图像融合方法，分别采用基于局部能量和改进拉普拉斯能量和的方法对低频和高频系数进行系数选择；②针对Curvelet变换的移变性缺陷，提出一种基于人眼视觉系统特性和循环平移Curvelet变换的图像融合新方法，在低频和高频的处理过程中，充分考虑人眼视觉特性；③基于非下采样Contourlet变换(NSCT)的多方向性和平移不变性，提出一种基于相位一致性和Log-Gabor能量的NSCT域医学图像融合方法；④基于医学图像的模糊特性和模糊熵理论，提出一种基于NSCT与二型模糊逻辑技术的多模态医学图像融合方法；⑤基于脉冲耦合神经网络(PCNN)模型，构建一种基于NSCT与邻域熵PCNN算法的多模态医学图像融合模型；⑥为克服多尺度变换域中图像融合子带系数选择的困难和克服空间域图像融合中的块效应问题，提出一种新的两步混合图像融合方法。本项目的研究为结构和功能医学图像融合提供了新的研究思路和方法，其结果提高了医学图像融合的质量和性能，对于多模态医学图像融合在临床诊断和医学研究上具有重要价值。就上述内容，本项目取得了一系列重要成果，共发表学术论文18篇，其中13篇被SCI收录，1篇被EI收录。