摘要
核医学领域PET/CT可以为胸腹部的肿瘤诊断提供影像支持。但由于患者在扫描过程中不自主的呼吸运动,会造成呼吸运动伪影。为了提高PET图像质量,文中提出了一种无监督的图像配准校正框架,该方法中通过三维卷积神经网络(3D-CNN)预测图像的配准域,再由空间变换网络(STN)对图像进行扭曲变换,实现对PET图像的伪影校正。实验结果表明,在仿真的PET几何体模和像素体模数据集上分别取得了82.12%和83.76%的相似性Dice值,证明了该方法的有效性。
PET/CT in the field of nuclear medicine can provide image support for the diagnosis of thoracic and abdominal tumors.However,due to the patient’s involuntary respiratory movement in the scanning process,it will cause respiratory motion artifacts.In order to improve the quality of PET image,this paper proposes an unsupervised image registration correction framework.In this method,the registration domain of the image is predicted by three-dimensional convolutional neural network(3D-CNN),and then the image is distorted by spatial transformation network(STN),so as to realize the artifact correction of PET image.The experiment results show that the registration method in this paper achieves 82.12%and 83.76%similarity dice values on the simulated pet geometric phantom and pixel phantom data sets,respectively,which proves the effectiveness of this method.
作者
侯一波
佘波
贺建峰
HOU Yi-bo;SHE Bo;HE Jian-feng(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China;Yunnan Key Laboratory of Smart City and Cyberspace Security,Yuxi 653100,Yunnan Province,China;PET-CT Center,First People’s Hospital of Yunnan,Kunming 650034,China)
出处
《信息技术》
2022年第12期12-18,23,共8页
Information Technology
基金
国家自然科学基金项目(82160347)
云南省智慧城市与网络空间安全重点实验室项目(202105AG070010)
云南省重大科技专项(202102AE090031)。
