High Quality Low-dose Cerebral Perfusion CT Imaging Based On Sinogram Restoration And Dictionary Learning

Acute ischemic stroke(AIS),known as acute cerebral infarction,is the most common cerebrovascular disease in clinical practice.This kind of circulatory disorder leads to the limitation or comprehensive neurological deficit syndrome with high incidence rate,which is the important reason for the death and disability of the aged in our country.Cerebral perfusion CT(PCT)is widely used in clinical practice for AIS examination,but continuous dynamic scan within the diseased region may increase the absorption of x-ray dose,even cause some genetic diseases.How to reduce the scanning dose and not influence the diagnostic image quality at the same time has become the problem need to be solved urgently in the CT field.At present,there are many methods to reduce the CT x-ray dose.It is a simple and cost-effective way to minimize millisecond-seconds(mAs)within the acceptable image quality in clinic.However,with the decrease of mAs,the number of photons collected by the detector drastically decreases,so that the projection data is contaminated by a large amount of photon noise,and the reconstructed image quality is seriously impaired.The image shows a lot of noise and artifacts.For this kind of low mAs technology,in order to obtain the final high-quality parameter maps,we can starte from the projection data to restore the low-dose noise polluted sinogram,or using iterative reconstruction to replace the traditional filtered back projection(FBP)for image reconstruction.We can also base on the PCT sequence images themselves,through various noise suppression means to obtain high quality PCT sequence images.This method,also known as image post-processing technology,commonly performed in various filter forms.We can even bring the regularization in the deconvolution step for quantitative parameter estimation to directly estimate clinical available diagnosis blood perfusion parameter maps.By reducing the mAs,a large amount of photon noise appears in the projection data.According to a large number of experimental data statistical analysis,CT projection data has a certain statistical characteristics.Based on the statistical characteristics,we can design the corresponding denoising model and decrease the noise from the source.On the other hand,in the initial reconstructed low-dose PCT sequence images,we can observe that the PCT sequence images have the same background information.This part with fixed anatomical structure remains basically unchanged during the contrast agent concentration changing over time,which means that there is a large amount of redundant information between sequence images.Thus,we can divide the PCT sequence images into background and enhancement parts.Observing the perfusion change(enhancement)of the CT values within a piece of tissue is more precisely a regional effect rather than a single pixel contribution,it may be more practical significance to deal with the image as a number of blocks.Based on the information above,this paper intends to use the first and second category methods,the main work summarized as follows:(1)A low-dose CT perfusion imaging method based on penalty weighted least-square(PWLS)projection data recovery is proposed.This method takes the statistical distribution characteristics of brain PCT projection data into account,and uses the statistical properties of projection data for model,uses PWLS method to recover the data,and utilize the Gauss-Seidel(GS)method for iterative solving.Moreover,adaptive weighting is introduced between the original projection data and the projection data after PWLS restoration,so that the projection data can be recovered better.The method has the following advantages:①It makes full use of the statistical distribution characteristics of projection data,and helps to remove the noise more deeply;②GS optimization algorithm is used to solve iteratively;③ According to the level of projection data noise,adaptive projection data weighting is introduced to make the reconstructed result more accurate.The reconstruction result from the present method shows a smaller noise degree compared with other filtered ones.(2)A low-dose dynamic cerebral perfusion imaging method based on couple dictionary learning(CDL)was proposed.This method deals the background information and the enhancement information from the PCT sequence data separately,and the K-SVD dictionary learning method was used to train the 2D background information and 3D enhanced information for restoring the low-dose PCT sequence,and then estimated the high-quality cerebral blood perfusion parameters from the sequence.The method has the following advantages:①Taking full account of the structural characteristics of the PCT sequence image,and distinguishing the background information from the enhanced information;②Considering the noise level of the low-dose PCT sequence image,the normal-dose PCT sequence was introduced as the prior information for the training of the three-dimensional enhanced information dictionary.By taking the prior and temporal information properly,we can restore the PCT series better compared with other DL-based methods.