High Quality Low-dose Brain CT Perfusion Imaging Method

Perfusion Imaging is an imaging method based on flow effect which obs-erves the micro movement of molecules, While MRI angiography observes the macroscopic flow of blood. CT perfusion imaging technology was first put f-orward by Miles in 1991. And he did some animal experiments and explored the clinical applications successively. In brief, CTP imaging can be described as a process that a few seconds after high pressure injection of the contrast agent, continuous dynamic scanning of selected slices of interseted in cine m-ode is performed to acquire the time-density curve (TDC) of each pixel, then estimate the cerebral hemodynamic perfusion parameters by using selected al-gorithm to acquire a digital matrix which is showed by pseudo color. The theoretical basis of CTP is the Radioactive tracer dilution principle of nuclear imaging and the central volume principle:blood flow= (blood volume)/(mean transition time).At present, manyfunctional imaging methods have been applied to cerebral blood flow perfusion, such as Xe-CT, perfusion MR,PET (Positron Emission Tomography), SPECT (Single Photon Emission Tomography), and CTP. Specifically, (1) Xe-CT makes use of inert gas as diffuse tracer and has an accuracy measurement which makes Xe-CT a trustworthy means of measurement. However, the method is cumbersome to operate, and acquired images have some bone artifacts and motion artifacts. At the same time, xenon gas also has some side effects. (2) Perfusion MR imaging has a strong advantage on the evaluation of cerebral blood flow, but the relationship bewteen the tissue enhancement and tissue blood flow is logarithmic but linear, which will affect the quantitative measurement of CBF and CBV directly. (3) Perfusion SPECT and PET imaging have solid imaging foundation and thorough researchment because of the early clinical application. But their main disadvantages are poor spatial resolution and unable to calculate absolute blood flow volume. PET can be used as the golden standard of CBF, CBV and MTT measurement. However, PET examination is expensive and is difficult to achieve clinical routine use. (4) CTP examination has the ability to obtain anatormy, hemodynamics and pathogeny information and has an advantage on without using radioisotope and xenon, high spatial resolution, accurate quantitative measurement, economical and practical, simple and convenient over other imaging methods.Because of its remarkable performance in density, spatial and temporal resolution and the realization of the three-dimensional cerebral hemodynamic parameters evaluation efficiently, rapidly, noninvasively and accurately, Brain Computed Tomography Perfusion Imaging (Brain-CTP) has achieved a great success in Acute Stroke patient (ASP) diagnosis. Brain-CTP imaging can obtain the hemodynamic parameters of ASP, offer perfusion image of the necrotic zone and penumbra, then it provide valuable image information for rapaid diagnosis of ASP of ’time is brain’. However, the Brain-CTP imaging protocols need continuous dynamic scanning of selected interested slices which make the patients receive more X-ray radiation dose compared to conventional CT examination. The dose security report of Brain-CTP examination announced by the Food and Drug Administration (FDA) in October 2009 points out that the X-ray dose received by the ischemic ASP in Brain-CTP examination is eight times more than standard one. Optimal control of X-ray radiation dose has become a key issue in field of CTP.There are many different methods to reduce X-ray radiation dose, the most typical one is by lowering the tube current, shortening the exposure time to reduce the milliampere second (mAs) of X-ray tube, and reducing the scanning sampling points which named sparse point sampling. Specifically, (1) increasing the time interval of the continuous scanning:in order to obtain high quality of cerebral perfusion parameter images, Brian-CTP examination usually adopts 1 second interval scanning mode (the time interval of the continuous scanning of the selected slices is 1 second), and the total scanning time is about 40 seconds. Admittedly, increasing scanning interval or reducing the total number of scanning can achieve the goal of reducing radiation doses of Brain-CTP examination, but the corresponding cerebral perfusion parameters will deviate from the truth. Latest research indicates that the maximum Brain-CTP scanning time interval meeting with the needs of the diagnosis is also related to the amount of the injected contrast medium, which makes reducing radiation dose by increasing scanning time interval exist many objections and need further study. (2) individualizing scan protocol:in order to reduce the X-ray radiation dose, the image doctor or technician automatically adjust exposure tube current according to the anatomical morphology of patients and the differences of individual viscera in CT scan which can reduce the X-ray radiation dose patient received and guarantee the quality of the image and then realize the individualization CT examination of optimizing radiation dose. (3) data undersampling:undersampling data in the Brain-CTP imaging is the projection data achieved by reducing the projection angle in a single scan. Brain-CTP image reconstruction based on undersampling data as an important research direction in the field of CT imaging has been carried out for many years, but most of them are based on C-arm CTP imaging. Because the undersampling data is no longer meet the requirement of accurate CT reconstruction, traditional analytical methods (such as filter back projection method) lose effectiveness which leads to severe artifacts (streak artifacts) in the image.Up to now, the X-ray radiation dose reduction methods used in clinical are primarily to reduce the tube current. This method introduces a large number of photons noise in the projection data inevitably which degrade the quality of reconstructed image severely and show up as serious noise and artifacts in the reconstructed image. Because the hemodynamic parameters are calculared from the Brain-CTP images and the existence of noise and artifacts will affect the accuracy of parameters measurement directly, Clinical misdiagnosis may occur.The existing methods for this type of low-dose imaging are mainly divided into three categories. (1) Filter the clinical low-dose Brain-CTP image directly to reduce the noise and artifacts. These methods belong to the image post-processing technology and have been widely studied. Generally, this achieved by introducing prior and the design of filter. (2) Establishing a model of data recovery according to the noise characteristics of projection data, then reconstructing the restored data by FBP (filter back projection). Because of the incorporation of the statistical features of projection data in the design of the filter, this kind of algorithm is effective and has a good uniformity for the reconstructed image. The noise of projection data are generally treated as gaussian or poisson noise. (3) Complete the statistic-based Brain-CTP iterative reconstruction according to Statistical regularity of the projection data. The statistical reconstruction algorithm can describe the CT system by the accurate measurement equation and is easy to introduce a priori information constraint, so it’s suitable for high quanlity low-dose CT image reconstruction.In general, majority current existing statistical iterative reconstruction algorithm of CT images can be used in low-doses Brain-CTP.With regard to the cine scanning mode of CTP, the acquired sequence images usually contain the same anatomical structures except for some anatomical changes due to involuntary movement of the patients. In other words, there are a lot of redundant information among the CTP sequence images. So we can ’split’ the CTP sequence images into the anatomical part (same for each time frame) and the part that the contrast medium changes over time. Therefore, how to use the separable feature of the CTP sequence images is the focus of this research. On the other hand, the pixel dimensions in a typical CTP image are much smaller than the tissue structure, Specifically the pixel spacing of our clinical data is 0.43 mm between the centers of adjacent rows and columns.In comparison, the tissue structure of the white matter and gray matter usually in the range of 10-50 pixels with relatively similar anatomical structure or functions. So changes in perfusion are regional effects rather than single pixel effects. Changes of the perfusion information in the same tissue should be the same or similar. Thus, how to make use of the regional characteristics of the perfusion information change is also the focus of this research.To sum up, the main work of this paper is as follow:(1) To propose a low-rank and total variation (TV) regularization based deonvolution algorithm. We assume that changes in perfusion are regional effects rather than single pixel effects. Within extended pixel neighborhoods the perfusion parameters will be constant or of low-variation, while it is also important to identify edges between different regions where tissues undergo perfusion changes, particularly deficit regions. In this paper, we punish regional temporal correlation of tissues by low-rank regularization, punish spatial correlation by TV regularization, and at the same time we introduce a novel variable splitting algorithm for the efficient to optimize the objective function. This algorithm has four major advantages:(1) we propose to regularize the impulse residue functions instead of the perfusion parameter maps; (2) the optimization is performed globally on the entire spatio-temporal data, instead of each patch individually; (3) there is no need of training data or the learning stage, and (4) our approach is able to compute all the common perfusion parameters, including CBF, CBV, MTT. Both digital phantom and clinical experiments show that the low-rank and total variation (TV) regularization based deonvolution algorithm proposed in this paper has strong robustness and can achieve the accurate calculation of the hemodynamic Parameters of low-dose CTP images.(2) To propose a sparse and low-rank matrix decomposition (SLMD) model for low-dose CTP image recovery. We perceive the enhanced images as a mixture of sparse matrix and low-rank matrix to explore the maximum temporal coherence of the spatial structure among phases. We assume that the density change of the contrast agent in the tissue can be regarded as a sequence of spatial images with different temporal sparse ’motion’ or ’change’ from a common ’background’. The sparse matrix corresponds to the time-varying component, which is often either approximately sparse itself or can be sparsified in the proper basis; the low-rank matrix stands for the background, which is stationary over time. Subsequently, an alternating direction method was adopted to optimize the associative objective function. In the results of clinical study, the results suggest the SLMD model proposed inthis paper can achieve the goal of high quality recovery of the low-dose CTP images.