Iterative reconstruction of low-dose CT based on differential sparse. (January 2023)

• Record Type: Journal Article
• Authors: Lu, Siyu; Yang, Bo; Xiao, Ye; Liu, Shan; Liu, Mingzhe; Yin, Lirong; Zheng, Wenfeng
• Is Part Of: Biomedical signal processing and control. Volume 79(2023)Part 2
• Publication Date: 2023-01
• Subjects: Low dose CT -- Sparse representation -- Sparse transform -- Iterative reconstruction -- Image processing; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104204
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Focuses on how to reduce the radiation dose of CT and ensure CT's imaging quality. Proposes a discriminative sparse transform iterative reconstruction algorithm inspired by the prior image compressed sensing reconstruction and the differential feature representation model. Data are low-dose 840 × 840 CT images obtained by simulating the fan beam scanning scene from the heart trunk phantom CT data. The results show that the discriminative sparse transform constraints can effectively introduce a priori image and reconstruct a better image effect and avoid the registration and matching problem of the reconstructed image caused by the difference of the prior image source. Abstract: The commonly used method to reduce the dose is to reduce the tube current. The number of photons received by the detector decreases, making the CT image obtained by analytical reconstruction full of speckle noise and strip artifacts. It interferes with the diagnosis and analysis of the disease. Therefore, how to reduce the radiation dose of CT and ensuring CT's imaging quality is an important research topic in the field of low-dose CT. This paper proposes a discriminative sparse transform iterative reconstruction algorithm inspired by the previous image compressed sensing reconstruction and the differential feature representation model. The global constraint term is used to constrain the consistency between the projected data to be reconstructed and the real projection data. The prior information constraint term constrains the reconstructed image close to the preceding image. This paper adds low-dose CT images obtained from image post-processing based on learning sparse transform to the prior information. Compared with the global constraints constructed only by learning sparse transform, the discriminative sparse transform constraints can effectively introduce a priori image and reconstruct a better image effect. Also, the improved algorithm's prior image avoids the dependence of the classical prior image compression sensing reconstruction and the differential feature representation model on the prior image and avoids the registration and matching problem of the reconstructed image caused by the difference of the prior image source.