Research And Application Of Megavoltage CT Image Denoising Algorithm Based On Discriminative Feature Representation

The Helical TomoTherapy(Helical TomoTherapy)system is the most advanced cancer radiation therapy device in world.It combines dose guided radiation therapy(Dose Guided Radiation Therapy,DGRT),image guided radiation therapy(Image Guided Radiation Therapy,IGRT)and intensity modulated radiation therapy(Intensity Modulated Radiation Therapy,IMRT).The megavoltage CT(Megavoltage CT,MVCT)images were acquired and reconstructed by the fan beam emerging from the 3.5 MV source of the TomoTherapy system and the CT detector array on the opposite side.MVCT images had occurred Compton Effect during imaging,with reduced atomic number dependence,soft tissue contrast and density resolution were poor,and doping noise,which brings difficulties to IGRT and adaptive radiation therapy(Adaptive Radiation Therapy,ART).In this paper we consider MVCT images as a low quality CT image(Low Quality CT,LQCT),the images are unclear at present and can be employed for patient position correction only,while image registration,tissue and organ delineation are limited.In this article,the research and application of MVCT image processing technology in helical Tomotherapy image guide radiotherapy is the main line,a traditional CT image denoising and enhancement method was introduced into MVCT image,on this bases a three-dimensional block matching and discriminative feature representation algorithm was proposed,and combined these two algorithms with parallel acceleration algorithm to implement accelerated processing of the algorithm.Finally,the MVCT image denoising and enhancement system was designed and developed on this foundation.In view of the unclear situation of MVCT images,an effective postprocessed method-Discriminative Feature Representation(DFR)algorithm was applied to MVCT images for denoising and enhancement processing.The MVCT images are regard as the superposition of the high quality CT images and the noise images,the DFR algorithm was performed to select the high quality feature atoms and noise atoms in discriminative dictionary respectively,these atoms were employed to realize the sparse representation of MVCT images,and finally the high quality atoms were separated to reconstruct high quality MVCT images,adjusting the window width and window level of denoised MVCT images to implement images enhancement.The experiment results show that the DFR algorithm has achieved better denoising and enhancement effect in application of MVCT images(the average CNR of before and processed image: 5.6456 vs.4.5225).This article proposes a Block Matching 3D and Discriminative Feature Representation(BM3D+DFR)algorithm.Firstly,the algorithm performed the 3D block matching algorithm to implement block of original MVCT images,selecting the image block as the reference block in search window,selecting the image blocks that are similar to reference block as the matching blocks,reference and matching blocks to be stacked 3D image blocks(group).Then,DFR algorithm was used to process stacked 3D blocks,the 3D contraction transform was employed to process the 3D images block that was in DFR algorithm further to attenuate the noise.Finally,image blocks were returned to their original location for reconstruction MVCT images,and image enhancement was implemented by adjusting window width and window level.The experiment results show that BM3D+DFR algorithm has better effect than simple DFR algorithm,and it implements the denoising and enhancement of MVCT images(the average CNR of the BM3D+DFR and the DFR and BM3 D method: 5.9569 vs.5.7575).The Matlab and OpenMP parallel accelerate algorithm combines above denoising and enhancement algorithm to implement MVCT image acceleration processing in this article.Parallel operations in Matlab and OpenMP parallel were instead of single-threaded image processing in the original algorithm.The examination results show,compare to the single-threaded image processing,Matlab parallel algorithm is increased by 2 times in terms of speed,and the OpenMP parallel image processing speed is increased by 5 times.Combine the proposed algorithm,design and implement denoising and enhancement system for MVCT images in this article.The system integrates the image denoising,enhancement and acceleration algorithms and visually displays the contrast effects and objective evaluation results of original and processed MVCT images.At the same time,the system also includes image delineation and zoom functions,which achieves clinical application needs.