| Medical image information is playing an important role in the diagnosis of human health and clinical disease in recent years,it can be obtained by a variety of technical means.Among them,Magnetic resonance imaging(MRI)is a common medical image acquisition technology at present.Because magnetic resonance(MR)image technology is a high-resolution medical imaging technology for human tissues and organs,it can multi-angle and multi-planar imaging of various parts of the human body,so it is broadly used in clinical diagnosis.Nevertheless,during the process of MR image acquisition,the random noise interference of Rician distribution will be introduced,the useful information of images will be reduced and cut down the accuracy and effectiveness of imaging,which will affect the correct diagnosis and treatment of clinicians directly.In addition,there are few researches on the noise removal of Rician noise at present,and the denoising technology is not yet successful.In this paper,in view of the problem of Rician noise in the process of medical magnetic resonance imaging acquisition,some new methods of MR image denoising are proposed.The main research contribution of this paper includes the following.High levels Rician noise exists in medical MR image,the dual-domain filtering image denoising algorithm removes Rician noise incompletely and has long running time.To address these problems,a novel fast denoising algorithm for medical MR image was proposed in this paper.Our method combined dual-domain filtering image denoising algorithm with guided image filtering.First,in order to cut down the running time of algorithm,the result of the guided image filtering was used as the original guiding image of the dual-domain filtering image denoising algorithm to reduce the number of iterations.Then,we researched the influence of weight coefficient on denoising in the dual domain filtering algorithm,the weighting coefficient of algorithm was improved.The new method combined the original weighting coefficient with the exponential kernel function to construct a new weighting coefficient so as to obtain better denoising results.The results demonstrated that the proposed new method can suppress medical MR image noise quickly and protect the details of the image.Meanwhile,our method achieved better peak signal-to-noise ratio(PSNR)and higher structural similarity(SSIM)compared with some excellent medical MR image denoising methods.Moreover,compared with the classical dual-domain filtering image denoising algorithm and the non-local mean algorithm,the running time of the improved algorithm was reduced by 1/3 and 1/2 respectively.Due to Rician noise exists in medical MR image,A novel deep residual learning denoising algorithm for medical MR image was proposed.First,the clean images are removed from the noisy observation images by residual learning strategy,and the reserved the noisy of images are used for model training.The algorithm combining with batch normalization(BN)to improve the denoising performance.Then,We compared two optimization algorithms in the deep residual learning respectively,the stochastic gradient descent(SGD)and adaptive moment estimation(Adam).the Adam was selected and used to find the minimum parameters of the loss function and solve the optimization problem of parameters.Moreover,the influence of image pixel size on the deep residual network was studied and the appropriate pixel size was chose.Besides,on the basis of the traditional deep residual network,the pooling layer can be added to quicken the training speed of the network,improve the performance of the network and retain the useful information.In addition,we evaluate quantitatively the quality of the reconstructed images by peak signal to noise ratio and structural similarity.Experimental results show that the proposed method can suppress Rician noise from the medcial MR image effectively,protect the details of the image and achieve favorable denoising performance comparing with the traditional image denoising techniques. |
PDF Full Text Request