基于双次梯度联合改进Criminisi算法去除超声图像中的人工标记并修复图像的可行性  

Feasibility of removing manual marks on ultrasonic image and repairing images based on double gradient combined with improved Criminisi algorithm

作　　者：张钒 奚谦逸 李奇轩 焦竹青[1] 倪昕晔 ZHANG Fan;XI Qianyi;LI Qixuan;JIAO Zhuqing;NI Xinye(School of Microelectronics and Control Engineering,Changzhou University,Changzhou 213164,China;Department of Radiotherapy,Changzhou Second People's Hospital Affiliated to Nanjing Medical University,Changzhou 213003,China;Medical Physics Research Center,Nanjing Medical University,Changzhou 213003,China;Jiangsu Medical Physical Engineering Research Center,Changzhou 213003,China;Changzhou Key Laboratory of Medical Physics,Changzhou 213003,China)

机构地区：[1]常州大学微电子与控制工程学院,江苏常州213164 [2]南京医科大学附属常州第二人民医院放疗科,江苏常州213003 [3]南京医科大学医学物理研究中心,江苏常州213003 [4]江苏省医学物理工程研究中心,江苏常州213003 [5]常州市医学物理重点实验室,江苏常州213003

出　　处：《中国医学影像技术》2023年第3期429-434,共6页Chinese Journal of Medical Imaging Technology

基　　金：江苏省重点研发计划社会发展项目(BE2022720);江苏省卫生健康委医学科研立项面上项目(M2020006)。

摘　　要：目的观察基于双次梯度联合改进Criminisi算法去除超声图像中人工标记并修复图像的可行性。方法选取30幅二维声像图,图中均包含十字、箭头和/或文字标记,以20幅无标记图像作为参考。算法由标记提取模块及图像恢复模块两个部分组成,前者采用双次梯度最大连通面积方法,后者采用改进重加权Criminisi算法;以峰值信噪比(PSNR)和结构相似性(SSIM)作为指标评价修复图像的质量。结果基于双次梯度联合改进Criminisi算法可准确检出超声图像中的人工标记并生成掩模,用于去除标记、恢复图像。相比无标记图像,30幅超声图像提取的标记掩模的平均检测精度和平均错误发现率分别为0.96和0.63,修复图像的平均PSNR及SSIM分别为46.78 dB和0.99。结论基于双次梯度联合改进Criminisi算法可有效去除超声图像中的人工标记并修复图像。Objective To observe the feasibility of removing artificial labels on ultrasonic images and repairing images based on double gradient combined with improved Criminisi algorithm.Methods Totally 30 two-dimensional ultrasound images containing markers of cross,arrow and text were selected,while 20 unmarked images were taken as references.The algorithm consisted of two parts,i.e.mark extraction module and image restoration module,the former was established using double gradient maximum connected area which could completely extract the label region and screen out the normal region,and the latter was designed with improved reweighted Criminisi algorithm for image restoration.The peak signal to noise ratio(PSNR)and structural similarity(SSIM)were taken as quality evaluation indexes.Results Compared with original unmarked images,Criminisi algorithm based on double gradient could accurately detect artificial marks on ultrasound images,remove various labels and repair images.The average detection accuracy and average false discovery rate of 30 ultrasound images was 0.96 and 0.63,respectively.The average PSNR and SSIM of images recovered by Criminisi algorithm was 46.78 dB and 0.99,respectively.Conclusion The artificial labels on ultrasonic images could be effectively removed and the images could be repaired based on double gradient combined with improved Criminisi algorithm.

关 键 词：超声图像 标记提取 图像修复 

分 类 号：R445.1[医药卫生—影像医学与核医学] TP751[医药卫生—诊断学]