Study On Advanced Medical Computed Tomography Reconstruction Based Or Self-similarity And Sparse Representation

Since the introduction of X-ray computed tomography (CT) in 1972, CT has been applied extensively in medical and industrial fields. However, there are some problems in current CT technology. On the one hand, as more and more CT scans are performed every year, it becomes more and more important to control and reduce the radiation dose. One of the simplest and cost-effective way is to lower the tube current (m.As) of the X-ray tube as low as possible under the current CT system. However, this could cause photon starvation and consequently make the reconstructed images degraded. On the other hand, traditional CT assumes that X-rays are monochromatic. In reality, X-rays of CT scanners are polychromatic. The spectrum of X-rays contains smooth and continuous curve and some sparks from characteristic k lines for certain atoms. When X-rays penetrate the human body, photons with low energy will be mostly absorbed with high-energy photons left, mak-ing the average energy higher, which is called the hard-beam effect. If we do not take the hard-beam effect into account, it will cause hard-beam artifacts in the reconstructed images. In order to solve these two problems, compa-nies of medical instruments, researchers in CT field and medical professionals make unremitting efforts.In the framework of low-dose CT, in order to improve the image qual-ity of low-dose CT, experts have done some research, including the filtered backprojection (FBP) reconstruction method based on projection restoration and direct image filtering algorithm. However, these two kinds of methods have some disadvantages. As for the FBP reconstruction method based on projection restoration, there are some stripe-like artifacts in the reconstruct-ed images. As for the direct image filtering algorithm, due to the ignorance of statistical properties of projection data, reconstructed images always lose some important details. Hence, we make use of the statistical properties of projection data to denoise adaptively, and combine denoising methods in pro-jection domain and image domain. Finally, we develop some more effective denoising algorithms to improve image quality for low-dose CT.In the framework of spectral CT, to solve the beam-hardening arti-facts problem, experts did a lot of research. There are mainly three kinds of methods, i.e. injecting iodine contrast material, iterative reconstruction algorithms based on hardware modification, and traditional iterative algo-rithms under the current CT system. But they did not solve this spectral reconstruction problem radically.In order to reduce the radiation dose and improve image quality of CT imaging, based on the self-similarity and sparsity of CT data, we develop some advanced medical CT reconstruction algorithms from the following five aspects to solve the low-dose CT imaging problem and spectral CT nonlinear reconstruction problem.(1) Low-dose CT imaging based on an adaptive nonlocal fil-tering in projection/image domain. In sinogram domain, in view of strong self-similarity contained in the special sinusoid-like strip data, we pro-pose an adaptive non-local filtering with smoothness parameters adjusted adaptively to the standard deviation of noisy sinogram data, which makes the method much more effective for noise reduction. In image domain, we develop a sinogram-induced NLM filtering. In order to get a better and more appropriate weight, the average weights are related to both the image directly FBP reconstructed from noisy sinogram data and the image FBP reconstructed from restored sinogram data. Simulation experiments show that, our proposed method by filtering in both projection and image do-mains, has a better performance in noise reduction and details preservation in reconstructed images.(2) Low-dose CT imaging based on an adaptive filtering with self-similarity. We further study low-dose CT imaging and get a better denoising effect than that of (1). In sinogram domain, first use an adaptive median filtering to remove the isolated data, and then use an adaptive nonlo-cal means filtering to deal with non-stationary gaussian noise. Different from (1), the median filtering here makes use of the local total variation and the maximum pixel value to determine the threshold, which makes the orienta-tion of isolated data more accurate. In addition, combined with the adaptive filtering based on the statistical properties of sinogram data, our method can preserve important features and high accuracy of the data in sinogram do-main. In image domain, we propose a NLM filtering, whose average weights are related to both the image directly FBP reconstructed from the noisy sinogram data and the image FBP reconstructed from the restored sinogram data in a framework of weighted average processing. This weighted average processing makes our proposed method more effective. All conclusions here are further validated in simulation experiments and quantification analyses.(3) Low-dose CT imaging based on an adaptive guided filter-ing. Different from the projection data of our previous algorithms (1) and (2), the projection data in this section contain more noise, which is corre-sponding to much lower dose CT. In view of low computing efficiency induced by matching similar points, using strong self-similarity contained in special sinusoid-like strip data in projection domain, we propose a more efficient adaptive guided filtering, which combines methods in projection domain and image domain. First, in projection domain, a median filtering followed by a XLM filtering adaptive to levels of noise is used to process the sinogram data. Then, the FBP algorithm is employed to reconstruct images from previous filtered sinogram data. Finally, in image domain, guided by the FBP recon-structed image from the NLM filtered sinogram, curvelet hard-thresholding preprocessed Yaroslavsky (YFcurvelet) filtering is used to process the recon-structed image from the median filtered sinogram to obtain the final result. Simulation experiments validate that our proposed adaptive guided filtering in both projection and image domains gives a better performance in noise reduction and details preservation. Compared with some related filtering methods, our method has better reconstruction results and computing effi-ciency.(4) A framelet-based iterative maximum-likelihood reconstruc-tion algorithm for spectral CT. Dual-energy or multi-energy CT can produce spectral information of an object. To realize dual-energy CT, it uses either two scans, one for lower-energy and the other for higher energy X-ray, or one scan using an energy-discriminative detector. Unlike these methods, we propose an algorithm that can obtain an image with spectral information from just one scan with a current energy-integrating detector. Colored CT images can be created by merging reconstructed images at multiple energy levels. Comparing a color image with a gray image at one monochromatic energy, we find that many indistinguishable regions in the gray image can be differentiated clearly in the color image, for the reason that they have d-ifferent colors. Using the reconstructed images at different energies, spectral curves of the attenuation coefficient μ(r, E) can be obtained and analyzed at any point in the object. These spectral curves are a valuable tool in diagnos-ing the nature of a tumor. In this context, no hardware of the CT machine needs to be changed or adjusted. Under a polychromatic acquisition mod-el and based on a sparse representation in a framelet system, our proposed algorithm can reconstruct energy-dependent linear attenuation coefficients. Thus we solve the nonlinear spectral CT reconstruction problem.(5) A spectral interior CT by a framelet-based reconstruction algorithm. Reducing radiation dose is an important goal in medical com-puted tomography (CT) field, for which interior tomography is an effec-tive approach. There have been many interior reconstruction algorithms for monochromatic CT, but in reality, X-ray sources are polychromatic. Using a polychromatic acquisition model and motivated by framelet-based image processing algorithms, we propose an interior reconstruction algorithm to obtain images with spectral information assuming only one scan with a cur-rent energy-integrating detector. That is to say, our algorithm does not need any CT hardware modification, which can reduce costs of patients. This is a new nonlinear iterative reconstruction method by minimizing a special functional under a polychromatic acquisition model, where the reconstructed attenuation coefficients are energy-dependent. Experimental results validate that our algorithm can effectively reduce the beam-hardening artifacts and metal artifacts. It also produces color overlays using several reconstructed images corresponding to different energies, which are very useful in tumor identification and quantification.Our proposed CT reconstruction algorithms do not need CT hardware modification, and can be directly applied in the current CT scanners, which can reduce the radiation dose and costs of hospitals and patients. Based on self-similarity of data in projection and image domains and the framelet sparse representation, we can get much more clear and high-resolution recon-structed images, which contain more anatomic information and can provide more typical information for medical diagnosis. In a word, our advanced medical CT reconstruction algorithms have a great value of application.