Metal Artifact Correction In Head Computed Tomography Based On Convolution Neural Network

With the development of medical technology,more and more patients choose to implement metal implant surgery to improve the dental problems in the mouth,and the postoperative evaluation of these patients has always brought great trouble to the clinical.When patients have dentures,metal clips and other foreign bodies in their mouth,their CT images will produce bright and dark stripes or scattering artifacts because of the existence of these high-density metal objects,which leads to the blurring of the edge of the tooth tissue structure,which makes it very difficult for clinical medical diagnosis.In serious cases,it may even cause misdiagnosis,which poses a great threat to the followup treatment of patients.Metal artifact correction has always been a difficult and hot issue in the field of medical image processing.In this paper,we try to use deep learning technology to build a model to correct metal artifacts,which can be used in clinical diagnosis after surgery.We propose a new artifact correction algorithm based on the convolution neural network(Homographic Adaptation Convolution Neural Network,HACNN)to correct metal artifacts in oral CT.Considering that the human oral structure is roughly the same,we try to achieve effective removal of metal artifacts in the tooth region of CT images by learning and training between different human head CT slices.Firstly,a 17-layer convolutional neural network is established as the framework of deep learning,and then the features of original image and reference image are extracted by VGG19 network as the input of the improved Contextual loss function.Finally,the network model is trained and adjusted for many times to optimize the experimental effect.In our study,experiments were conducted on real clinical data sets and synthetic-simulation data sets to discuss the feasibility and correction ability of the model.In this paper,the deep CNN is developed and studied.The model can effectively remove the metal artifacts of head CT.The corrected CT reconstruction image not only does not produce new artifacts,but also does not cause blurring,and the disease details in the original CT image are well preserved.Peak signal to noise ratio(PSNR),structural similarity(SSIM),threshold segmentation,Dice coefficient and 3D reconstruction were used to evaluate the correction effect.Compared with other similar algorithms,the proposed method is proved to be more effective in correcting metal artifacts and preserving details.