基于动态规划拉普拉斯超声图像融合算法研究  

Research on Laplacian Ultrasonic Image Fusion Algorithm Based on Dynamic Programming

作　　者：陆正大 成定胜[1] 马宪礼[1] 唐局 薛谭 杨彬[1] 仲辉[1] 鲁军[2] LU Zhengda;CHENG Dingsheng;MA Xianli;TANG Ju;XUE Tan;YANG Bin;ZHONG Hui;LU Jun(Department of Medical Engineering,North Jiangsu People’s Hospital,Yangzhou Jiangsu 225009,China;Biobank,North Jiangsu People’s Hospital,Yangzhou Jiangsu 225009,China)

机构地区：[1]江苏省苏北人民医院医学工程处,江苏扬州225009 [2]江苏省苏北人民医院生物样本库,江苏扬州225009

出　　处：《中国医疗设备》2023年第11期54-59,共6页China Medical Devices

基　　金：苏北人民医院院内立项项目(YYGL202214)。

摘　　要：目的探究基于动态规划拉普拉斯超声(Ultrasound,US)图像融合算法图像辅助放疗靶区勾画的效果。方法基于动态规划思想获取US待融合图像重叠区域内邻域像素点相似度最高而色值差异最小的缝合线掩码权重图像,利用5层高斯金字塔进行多分辨率融合。提出一种基于动态规划下最佳缝合线拉普拉斯金字塔融合,通过在图像的重叠区域寻找最佳接缝来消除动态融合场景中的鬼影和重影。比较不同US图像融合方法的融合效果,同时比较全视图US和拼接US图像所描绘的膀胱区域与CT图像中的膀胱区域,验证本研究提出的融合算法应用于放疗中辅助CT勾画靶区的可行性。结果本文提出的基于动态规划最佳缝合线结合拉普拉斯金字塔融合算法显示了良好的性能,可从不同尺度将图像典型特征进行分解,并将图像细节信息保留到不同的塔型分解层上,还原US图像拼接区域的真实信息。全视图US和拼接US图像与CT比较,在x轴和y轴方向上的平均勾画误差差异均无统计学意义(P>0.05)。结论基于动态规划的最佳缝合线拉普拉斯算法获得了最佳的融合效果,能够最大程度保留图像的细节信息,且拼接US能够辅助CT完成靶区勾画。Objective To explore the effect of image-assisted radiotherapy target delineation based on dynamic programming Laplace ultrasound(US)image fusion algorithm.Methods Based on dynamic programming,the mask weight image of suture line with the highest pixel similarity and the smallest color difference in the overlapping region of the image to be fused was obtained,and the 5-layer Gaussian pyramid was used for multi-resolution fusion.A Laplacian pyramid fusion of optimal sutures based on dynamic programming was proposed to eliminate ghosts and ghouls in dynamic fusion scenes by searching for optimal seams in overlapping regions of images.The fusion effect of different US image fusion methods was compared,and the bladder region depicted in fullview US and splicing US images was compared with the bladder region in CT images,so as to verify the feasibility of applying the fusion algorithm proposed in this study to assist CT in mapping the target region in radiotherapy.Results The optimal sutures based on dynamic programming combined with Laplacian pyramid fusion algorithm proposed in this paper showed good performance.It could decompose typical features of images from different scales,retain image details on different tower decomposition layers,and restore the real information of the US image stitching region.Compared with CT,there was no significant difference in the average sketching error in the x-axis and y-axis directions between full-view US and spliced US images(P>0.05).Conclusion The optimal suture Laplacian algorithm based on dynamic programming can obtain the best stitching effect,and can retain the maximum details of the image.

关 键 词：动态规划 超声 高斯金字塔 图像融合 拉普拉斯 

分 类 号：R197.39[医药卫生—卫生事业管理]