Research On Low-dose X-ray CT Reconstruction Algorithm Based On Sparse Projection

As a high-performance non-invasive diagnostic technology,the X-ray computed tomography(CT)technology plays a crucial role in assisting clinical diagnosis.However,with the increasing use of CT technology,the hazards to public health caused by CT’s relatively high radiation doses have been paid more and more attention.Nowadays,the urgent need for CT development has changed to reduce radiation dose.Sparse angle scan is a more important measure to speed up data collection and reduce radiation dose,but incomplete projection data will reconstruct a severely degraded image.Therefore,the study of sparse reconstruction algorithms that can improve image quality has important significance and practical value for promoting the wider application of low-dose X-ray CT technology.This dissertation focuses on the improvement and optimization of the sparse reconstruction algorithm in two aspects.Major research works are as follows:(1)Based on structure tensor,a directional adaptive algorithm was proposed for sparse projection reconstruction.The directional total variation(DTV)model is an anisotropic model that effectively characterizes edge features in a single direction.On this basis,an anisotropic total variation model with directional adaptability was constructed by combining the unique advantages of the structure tensor.The new model utilized the structure tensor to realize the adaptive setting of direction and weight parameters,so as to provide different directions and degrees of diffusion depending on whether it is in a flat area or an edge area.Then the new model was incorporated into the iterative reconstruction algorithm architecture to achieve sparse projection reconstruction.The performance of the new algorithm was verified through experiments,and the reconstruction results of different sparse angle projections were compared and analyzed.Both subjective visual effects and objective evaluation results show that the new algorithm can achieve satisfactory results in streak artifacts suppression and edge contours preservation.(2)Based on the different characteristics of different sinogram regions,a sparse projection reconstruction algorithm for sinogram divisional inpainting was proposed.The proposed algorithm took projected sinogram by sparse sampling as the processing object and divided the target sinogram into two categories according to gray entropy.The dictionary learning method was used to repair the boundary area of the sinogram,which has strong sparsity.When repairing the inner area of the sinogram,a joint repair model was constructed,which contains a low-rank penalty in the sparse representation model.This model took into account both local sparsity and non-local similarity in the repair process,so it can further preserve the internal structure of the sinogram.The new algorithm used K-singular value decomposition(K-SVD)algorithm to train two dictionaries,which act on the two parts of the repair process.Finally,a complete sinogram was formed and then reconstructed by filtered back projection(FBP)to obtain a sparse reconstructed image with improved quality.The performance of the algorithm was verified in both the simulated head model and the actual pelvis model.Additionaly,the repair and reconstruction effects under different projection loss rates were compared and analyzed.The results show that the new algorithm has advantages in sinogram inpainting,and can achieve a better balance between streak artifacts suppression and structure blurring improvement.