摘要
多模态医学图像融合已成为有效结合正常组织结构和异常改变信息、提高医学诊断效率的强有力辅助技术.针对空域融合技术在处理图像细节丢失、光谱退化等方面的缺陷,提出一种在联合双边滤波器(JBF)域内实现图像结构与细节信息增强的双通道频域多模态医学图像融合方法.该方法将源图像分解为结构和能量2个通道,分别处理图像纹理细节信息和边缘强度信息.在结构通道中,通过改进梯度能量得到局部梯度能量算子,进一步提升融合图像对小尺度细节信息的表达能力和对噪声的鲁棒性;在能量通道中,利用非下采样轮廓波变换提高模型的多方向多尺度特性,并提出一种局部熵细节增强算子和脉冲耦合神经网络结合的高频子带处理框架,达到增强能量通道中结构信息和细节信息的效果.在Atlas公开数据集上,与基于MST、稀疏表示、PCNN以及JBF的6种具有代表性的频域融合方法进行对比及消融实验的结果表明,所提方法的融合图像与源图像相似度提升35.0%,空间频率提升16.2%,边缘保持度提升12.5%,对比度提升11.2%;并在视觉方面有较好的效果,明显优于对比方法.
Multi-modal medical image fusion has become a strong auxiliary technique that effectively combines normal tissue structure and abnormal alteration information to improve the efficiency of medical diagnosis.To address the shortcomings of space-domain fusion in dealing with image detail loss and spectral degradation,a two-channel frequency-domain multimodal medical image fusion method is proposed in the JBF domain of joint bilateral filter.This method decomposes the source image into two parts,structure and energy channels,to process the image texture detail information and edge intensity information,respectively.In the structure channel,the local gradient energy operator is obtained by improving the gradient energy,which further enhances the representation ability of small-scale detail information and the robustness to noise.In the energy channel,non-subsampled contourlet transform improves the multi-directional and multi-scale characteristics of the model.Meanwhile,the high-frequency sub-band processing framework that combining the local entropy detail enhancement operator and pulse coupled neural network is proposed to enhance the structural and detailed information in the energy channel.Experiments conducted on public dataset Atlas were compared with six representative frequency domain baselines,including methods based on MST,sparse representation,PCNN and JBF,etc.The results show that the similarity between the fused image and the source image is improved by 35.0%,the spatial frequency,edge intensity,and contrast ratio of fused images are improved by 16.2%,12.5%and 11.2%,respectively.Meanwhile,the visual effect obtained by this method is also significantly better than others.
作者
朱积成
刘慧
李珊珊
李攀
张彩明
Zhu Jicheng;Liu Hui;Li Shanshan;Li Pan;Zhang Caiming(School of Computer Science and Technology,Shandong University of Finance and Economics,Jinan 250014;Shandong Key Laboratory of Digital Media Technology,Jinan 250014;The Second Affiliated Hospital of Chongqing Medical University,Chongqing 400010;School of Software,Shandong University,Jinan 250101)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2024年第6期857-874,共18页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(62072274)
山东省科技成果转移转化项目(2021LYXZ011)
重庆市科技局鲁渝科技协作项目(cstc2021jscx-lyjsAX0003).
