基于U-net的双能血管减影算法

A dual energy subtraction algorithm based on U-net

传统的数字血管造影(Digital Subtraction Angiography,USA)技术是通过在病人体内注射造影剂前后两次曝光来实现的,其劣势是两次曝光间存在时间差,被摄物体运动产生位移,导致减影图像出现运动伪影。为解决这一问题,本文提出了一种使用双层平板探测器,仅通过注射造影剂后一次曝光实现双能量减影的方法。该方法使用高频分量图作为跳接结构图像的U-net网络,训练集为不含碘液的高低能头模图像,输入高能图像并以低能图像作为真值拟合;测试集为含碘液的高低能头模图像,输入高能图像后碘液区域因未经训练,输出的预测图像与低能图像相减,消去骨头、软组织等背景的数值,实现最终的血管剪影。最后,本文在锥形束CBCT成像平台上,设计了相关实验和评价指标,验证分析了上述U-net方法实现双能DSA剪影的可行性和性能,相较于传统多项式拟合方法在PSNR和CNR等指标均有显著提升。

The traditional digital subtraction angiography(DSA)is realized by two exposures before and after the injection of contrast agent to patients.Its disadvantage is that there is a time difference between two exposures,which causes motion artifacts in the subtraction image.In order to solve this problem,this paper proposes a method of dual-energy subtraction using a double-layer flat-panel detector with one exposure only after injection of contrast agent.We use U-net which contains high-frequency component images as shortcut connections.The training set is the high and low energy images of a head phantom without iodine solution.We input the high energy images and use the low energy images as the labels to fit.The test set is the high and low energy images of the head phantom containing the iodine solution.We input the high energy image and output the predicted image.Background values such as bones and soft tissues is eliminated by subtraction of low-energy images.The final blood vessel silhouette is achieved because the iodine area is not trained.Ultimately,we design related experiments and evaluation indicators on the cone-beam CBCT imaging platform to verify and analyze the feasibility,as well as the performance of the above U-net method to achieve dual-energy DSA.Compared with the traditional polynomial fitting method,the PSNR and CNR indicators are significantly improved for the proposed method.