三维CT层间图像超分辨重建与修复

CT图像在疾病诊断、癌症精准放疗阶段发挥着重要的作用.然而,受X射线剂量的限制以及现有医疗设备等因素的影响,三维CT图像成像采样间距较大,层间分辨率远低于层内分辨率.为提高CT图像的层间分辨率,避免因相邻层CT图像之间存在较大差异对超分辨率重建造成干扰,提出一种在相邻CT图像序列切片间利用上下层的配准信息插值出一个新的中间切片,并对插值得到的中间切片进行修复的方法.首先,利用CLG-TV光流估计模型配准算法估计相邻2幅CT图像之间的像素运动,得到一个稠密的光流场,从而获得2个连续切片之间精确的像素对应关系,并依据新切片位置对速度矢量进行缩放;然后,利用计算出的速度场在连续2幅图像之间生成初始插值图像.由于相邻切片之间的像素难以一一对应,插值后的图像通常存在像素缺失现象.最后利用序列图像的帧间非局部自相似性,通过求解最优化问题以修复插值图像中像素丢失的区域.在DIR-Lab实验室和山东省医学影像学研究所提供的数据集上的实验结果表明,与其他经典方法相比,文中方法能够生成高质量的中间CT切片,在定量和定性上提高了层间分辨率.

Super-Resolution Reconstruction and Inpainting of Inter-Layer Image in 3D CT

CT image plays an important role in the diagnosis of diseases and precise radiotherapy of cancer.However,due to the limitation of X-ray dose and the existing medical equipment,the sampling interval of 3D CT image is large,which results in the much lower inter-layer resolution of 3D CT images than the intra-layer resolution.In order to improve the inter-layer resolution and avoid the interference with super-resolution reconstruction from the large difference between adjacent layers,this paper proposes a new model for reconstructing an intermediate slice that combines the registration information from adjacent CT sequence,and then repairer the new slice via non-local self-similarity features.First,CLG-TV optical flow estimation model is used to estimate the motion between two adjacent CT slices,for obtaining a dense optical flow field.The model constructs the precise pixel correspondence between two consecutive slices,which can scale the velocity vector according to the position of new slice.Then,an initial interpolated image based on the calculated velocity field is generated.Since it is difficult to guarantee the one-to-one pixel correspondence between original adjacent slices,there generally exists the pixel missing phenomenon in interpolated images.Finally,non-local self-similarity among different frames is used to inpaint the missing regions in the interpolated image.The experimental results on the data sets provided by DIR-Lab laboratory and Shandong Medical Imaging Research Institute show that,compared with other classic methods,this method can generate high quality intermediate CT slices,and improve the inter-layer resolution quantitatively and qualitatively.