Research On Cone-beam CT Image Quality Enhancement

As one modern auxiliary medical and diagnostic tool,the cone-beam computed tomography(CBCT)can obtain the internal information of objects under the non-contact and non-destructive conditions.In the process of precise radiotherapy,the high-quality CBCT images are needed to obtain accurate target location.However,in the reconstruction process,the image quality is degraded by the geometric parameter error of each machine caused by some unfavorable factors,such as the imaging element,the component assembly accuracy,etc.Moreover,a certain level of noise is brought into the acquired images,due to the rotation of the X-ray source and the performance of the parallel plate detector.To address these issues,the following two works have been carried out in this thesis.(1)A CBCT geometric calibration algorithm is proposed based on the nonlinear estimation.First,a specific calibration phantom is designed to obtain the projection images in different directions and detect the feature points.Then,the corresponding coordinate system is established in terms of the geometric relationship of the CBCT imaging system.The relevant parameters are calculated by using the geometric relationship between the calibration phantom and the projected image.Further,the non-linear estimation is performed by using the detected feature points as samples,to improve the accuracy of calibration geometric parameters.Finally,the geometric parameters of the CBCT imaging system are calculated by combining the geometric relationships of different coordinating systems and the known parameters.The experimental results show that,the introduction to non-linear estimation can both improve the accuracy and reduce the requirements for the positioning of the calibration phantom before calibration.(2)An image denoising method is proposed based on the weighted sparse residual.First,the correlation coefficients of the image blocks are found according to the self-similarity of the target image.As the noise in the real image is highly random,the reference weight estimation is performed in each iteration.In addition,in order to introduce the self-similarity of image blocks in the sparse representation,a noise reduction method of sparse residuals is applied to the grouped image blocks.Finally,the image is reconstructed after the noise is reduced.The experimental results show that the introduction of noise weight can solve the problem of randomness of noise standard deviation without increasing the computational complexity.The proposed method is more suitable for real image noise reduction processing.