太原理工大学学报2025，Vol.56Issue(5)：839-854,16.DOI:10.16355/j.tyut.1007-9432.20240593
TLIR:基于双层迭代细化模型的有限角CT重建
TLIR:Limited-angle CT Reconstruction Based on a Two-layer Iterative Refinement Model
摘要
Abstract
[Purposes]The limited angle reconstruction is a typical discomfort problem in com-puted tomography(CT).In practice,owning to the limited scanning angles available for fixed scanning targets and the patients'ability to withstand radiation,complete projection data are usually unavail-able.Besides,images reconstructed by traditional analytical iterative methods show severe structural distortions and tilt artifacts.[Methods]To address the above problems,a Two-Layer Iterative Re-finement Model(TLIR)was proposed to recover the structural details of the missing parts in the limited-angle CT images and reconstruct high-quality CT images.Specifically,the denoising diffu-sion probabilistic model was modified to adapt it to conditional image generation for the image domain restoration problem,and the model output was started from a finite-angle CT image with noise alias-ing,and iteratively refined by using residuals trained in U-Net at various noise levels.In addition,since the deep learning model may corrupt the sampled portion of the sinusoidal data during the infer-ence process,a learnable data fidelity module called DSEM was proposed to recover the sinusoidal data tampered by the depth model.The two modules were executed alternately to form the TLIR model.The two-layer iterative structure also made the network more robust during the training and in-ference process.[Results]TLIR shows strong reconstruction performance under both 90°and 120° finite-angle sampling,with an average improvement of 2.000 9 dB and 2.5 dB,respectively,over the existing advanced methods in terms of peak signal-to-noise ratio,proving the validity of the present model.关键词
计算机断层扫描(CT)/迭代重建/有限角度CT重建/扩散模型/数据保真约束Key words
computed tomography(CT)/iterative reconstruction/limited-angles CT(LACT)re-construction/diffusion model/data fidelity constraints分类
信息技术与安全科学引用本文复制引用
李青,王涛,强彦,张斌,武炜..TLIR:基于双层迭代细化模型的有限角CT重建[J].太原理工大学学报,2025,56(5):839-854,16.基金项目
国家资助博士后研究人员计划(GZC20241586) （GZC20241586）
国家自然科学基金项目(62376183) （62376183）
山西省基础研究计划资助项目(202403021212184) （202403021212184）
山西省高等学校科技创新项目(2024L181) （2024L181）
