分割图像辅助的自监督医学图像配准  

Segmented image assisted self-supervised medical image registration

作　　者：李宗民[1] 王群[1] 李泫廷 杨超智 LI Zong-min;WANG Qun;LI Xuan-ting;YANG Chao-zhi(College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《计算机工程与设计》2025年第3期857-863,共7页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2019YFF0301800);国家自然科学基金项目(61379106);山东省自然科学基金项目(ZR2013FM036、ZR2015FM011)。

摘　　要：为解决可形变医学图像配准工作更关注于像素/体素层级的配准,忽略了扫描图像中的结构信息的问题,采用多结构特征提取模块(multi-structure feature extraction,MSFE)。以形变图像的分割图像和固定图像的分割图像作为输入,提取结构信息并反馈给配准神经网络。在此基础上,在配准任务常用的损失函数中加入与MSFE模块搭配的损失函数与分割任务中常用的损失函数以辅助配准。所提方法只参与配准网络的训练阶段,不参与测试阶段,不会增加实际配准时所需的时间。在OASIS Sample Data数据集上的实验验证了方法的有效性。To solve the problem that the deformable registration of medical images focuses more on the pixel/voxel level but ignores the structural information in the scanned images,the multi-structure feature extraction(MSFE) module was adopted.The segmented image of the deformed image and the segmented image of the fixed image were taken as input,the structural information was extracted and it was fed back to the registration neural network.The loss function paired with the MSFE module and the loss function commonly used in the segmentation tasks were added to the loss function commonly used in registration tasks to assist with registration.The proposed method is only involved in the training stages of the registration network but not in the testing stages,which means that it does not increase the time required for the actual registration.Results of experiments on the OASIS Sample Data dataset verify the effectiveness of the method.

关 键 词：深度学习 自监督 单模态 可形变配准 医学图像 医学图像配准 分割图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]