【摘要】 目的：为了解决人体颈椎CT序列图像横向点长度和纵向点长度不匹配带来的矢状面和冠状面结构变形问题，提出一种基于CT图像双向光流计算网络和CT图像双向光流增强网络的层间插值图像生成算法。方法：首先基于残差密集网络设计CT图像双向光流计算网络，计算上下层图像之间的双向光流；其次，利用CoTBlock模块改进UNet网络设计CT图像双向光流增强网络，提高双向光流精度并生成最终的层间插值图像。为验证算法的有效性，进行对比实验和消融实验验证。结果：提出的算法可以生成较高质量的层间插值图像，且在计算速度和模型参数量上有更好的表现。结论：该算法生成的插值图像毛刺伪影少、质量好，且经过层间插值后的人体颈椎CT序列图像更贴近人体颈椎的真实情况，有利于颈椎形态结构的研究和分析。更多还原

【Abstract】 Objective To propose an interlayer interpolation image generation method based on bidirectional optical flow calculation network and enhancement network for CT images so as to solve the problem of structural deformation in the sagittal and coronal planes caused by the mismatch between the transverse and longitudinal point lengths of human cervical spine CT image sequences. Methods Firstly, a bi-directional optical flow calculation network for CT images was designed based on the residual dense network to calculate the bi-directional optical flow between the upper and lower images;secondly, a bi-directional optical flow enhancement network for CT images was developed by improving UNet network using a CoTBlock module so as to increase the bi-directional optical flow accuracy and form interlayer interpolation images.Comparison and ablation experiments were carried out to verify the effectiveness of the proposed algorithm. Results The method proposed could generate high-quality interlayer interpolation images and gained advantages in computational speed and number of model parameters. Conclusion The interpolation images generated by this method have few burr artifacts and high quality, and the CT image sequences of human cervical spine after interlayer interpolation are close to the real situation of human cervical spine, which facilitate the study and analysis of the morphological structure of cervical spine. [Chinese Medical Equipment Journal,2023,44(1):27-34]更多还原