High Quality Low-Dose Computed Tomography Reconstruction Based On Wavelet-Transform And Non-Local Weights Prior

Ever since Roentgen discovered x-rays in 1895, X-ray has been widely used for the human examination, as the basis of disease diagnosis. Computed Tomography (CT) appeared in 1970s, combining the rapid developing modern computer technology and the X-ray technology. After nearly forty years'development, CT entered into a new stage, with the demand of different postures scan (head-lung-body) in 1970s, and the evolution of the scanning speed (2s-1s-0.75s) in 1990s, and included the strengthening of the scanning levels (2 slices-4 slices-16 slices-64 slices-128 slices, etc), and the emergence of dual-source,4D dual-source, and dazzle speed dual-source. CT's development achieved low radiation, high resolution, large scanning range, fast scanning speed, high quality images and precise stereo graphic. CT has been playing a more and more important role in clinical medicine, also has laid a good foundation for clinical medicine to higher development.With the rapid development of modern medicine, CT technology has become one of the core technologies in medical imageology. Currently, CT is at the peak of development and the new technologies has reexpanded its territory of applications. With the emergence of multi-slice CT and dual-source CT, some difficult techniques, such as the dynamic examination in heart and lungs and major organ perfusion, have been applied in clinical. Many new CT equipments used higher X-ray dose in order to get high quality medical images. However, high dose limits many aspects of its clinical applications, such as the high-risk group lung cancer sieves examines, the follow-up visit about the lung cancer after treatment and the chemotherapy, gravid, infants and the child lungs CT inspection. To lower the radiation dose, low-dose protocol has been used clinically as an alternative for above situations, but with the quantum noise in low-dose CT scanning, the quality of image decreases, affecting diagnostic accuracy.There exists certain noise in CT images. Especially when the radiation dose decreases, the image noise will increase dramatically. When the radiation dose decreases to certain level, with the increased noise, the images will not be able to meet the clinical requirements. To this end, the World Health Organization (WHO), the International Commission on Radiological Protection (ICRP) and the International Organization of Medical Physics (IOMP) developed a medical exposure quality assurance and quality control standards, and advocates that it should follow the X-ray diagnosis of the legitimacy of practices and optimization of protection principle (As Low As Reasonable Achievable, ALARA), with a minimum of cost and radiation dose to obtain the best diagnostic results, which provided a broad space for the low-dose CT technology application and research. Rational use of low-dose recommended that it should minimize radiation dose as low as possible for patients in meeting the requirements of clinical diagnosis.Compared to the traditional standard-dose scan, low-dose CT, which has vital significance in clinical application, has the following advantages:(1) Reduce the subjects'radiation hazards, improve image quality.Low-dose CT can reduce the subjects'X-radiation dose, and eliminates partial patient of X-ray fears, which can be used for large-scale population census and pregnant women, children's diagnostic tests. It can not only satisfy the diagnosis's needs, but also notice the forward radiation effect which may cause the adverse consequences.(2) Reduce the tubes and detectors'damages, Save CT operation costs.Low dose CT techniques can extend the life of the tube to reduce the CT operating costs, thereby reducing the diagnosis cost of patient.Reasonable application of the low-dose CT technology to gain a CT image that can meet the clinical requirements is the important issue in low-dose CT high quality reconstruction. The comparison of quality or diagnostic acceptance rate between low-dose CT and standard-dose CT images is a main aspect of the feasibility of application. Currently, the low-dose CT technology research mainly adds conventional noises into CT image through computer software to simulate low-dose CT. This can avoid directly test low-dose on patients. Low-dose CT on image quality assessment is still at trial stage, and most of the studies show that the difference between the standard-dose and low-dose in the detection of the disease has no statistical significance. Therefore, it gives a broader space for low-dose CT's development in clinical, "Green CT" gets more and more attention.Considering the redundancy between the previous standard-dose CT images and low-dose CT images, we introduce a new method to restore an image though standard-dose image as the weight of prior information, rather than low-dose images itself. Taking into account the redundancy between different slices of anatomical information, we used weight prior information in 3D low-dose CT image sequence restoration. Because of the 3D NLM's filtering capacity, GPU's strong parallel computing power, and the fact that the NLM algorithm is suitable for parallel processing, we used GPU as hardware acceleration. As the standard-dose images includes more and better redundant information, so our method gets a better effect than conventional NLM in low-dose images.At the same time, based on the analysis of the property of the low-dose CT projection noise, and according to the good local quality in time domain and frequency domain of wavelet transform, we use the stable Anscome variance transformation formula, and then the efficient BM3D algorithm is performed separately in the scale coefficient and the wavelet coefficients in frequency domain. Finally, the reconstruction image is achieved by the classical filtered back projection (FBP) method after inverse Anscombe transform and inverse Haar-wavelet. Unlike direct BM3D filtering method in the projection domain, our method uses wavelet transform signal adaptive, filter the wavelet decomposition's low and high frequency coefficients separately. Because the wavelet transform has a good time-frequency localization quality, and wavelet transform is linear transform, Gaussian noise through wavelet transform will still satisfy the Gaussian distribution, evenly distributed throughout the space, but mainly concentrated in the high frequency region. Therefore, filtering the decomposition of the wavelet coefficients and scale coefficients of the filter can achieve better results.According to the redundancy between the previous standard-dose CT images and low-dose CT images and the property of the low-dose CT projection noise, we propose two new high quality low-dose CT reconstruction methods:(1) According to the redundancy between the previous standard-dose CT images and low-dose CT images, this paper propose a non-local weights prior and GPU acceleration based 3D low-dose CT images reconstructions. The repeated CT scans are often needed in clinic such as tumor surveillance and radiotherapy planning. The radiation dosage in CT examinations has been widespread concerned by people. Due to the degradation of low-dose computed tomography (CT) image quality, a new image reconstruction method is proposed for low-dose CT images. The new method exploits the redundancy of information in previous normal-dose scan images to optimize non-local filtering. As the anatomical structural information of the two scanned images maybe inconsistent, the 3D low-dose images and the previous 3D normal-dose images are registered, then the NLM weights prior are also constructed. With the drawback of computational burden about 3D NLM filter, GPU is used to accelerate the reconstruction. Clinical 3D CT images data experiment results demonstrate that this method achieves between structure and details of low-dose CT image with the noise suppression, the efficiency of the algorithm has also been increased dramatically.(2) According to the property of the low-dose CT projection noise, this paper propose a wavelet-transform based low-dose CT projection BM3D Filtering (Block-Matching and 3D filtering) technique. For the proposed method, first, low-dose projection data is decomposed by Haar-wavelet transform after Anscombe transform. And then the efficient BM3D algorithm is performed separately in the scale coefficient and the wavelet coefficients. Next, the filtered coefficients are transformed orderly by inverse Anscombe transform and inverse Haar-wavelet transform. Finally, the reconstruction image is achieved by the classical filtered back projection (FBP) method. Simulated and clinical data experimental results demonstrate that the proposed method can do well in noise suppression and detail preservation of the reconstructed image.