Research On Automatic CT Image Processing Technology In Navigation Of Knee Replacement Surgery

Severe osteoarthritis can cause joint pain and impaired motor function,causing permanent damage to patients.Knee arthritis is the most common type of osteodegenerative joint disease.Knee replacement,as the most effective treatment for knee arthritis,has shown a trend from manual surgery to robot-assisted navigation surgery.Existing robot-assisted navigation knee replacement surgery requires the extraction of knee bone pixels before surgery for three-dimensional reconstruction of femur and tibia.The reconstructed model was marked with medical feature points of the knee joint for intraoperative registration to achieve surgical navigation.However,all the above operations are realized based on manual labeling.Manual segmentation of a set of knee CT images takes more than 40 minutes,and manual extraction of feature points depends on the doctor’s operation proficiency,which normally takes more than 3 minutes,and the preoperative preparation time is too long and the intelligence level is low.Therefore,this paper studies the automation of CT image processing in order to improve accuracy and efficiency.The research is mainly carried out in the following aspects:Firstly,in view of the characteristics of low data volume,high noise and high resolution of knee CT images,a high resolution generative adversance network based on gradient penalty was proposed,and the effetive data volume was increased by fitting real images with random noise.Deep convolutional neural network is used to learn the deep features of real images to increase the diversity of data.On this basis,introducing residual module generator and discrimination in the,in the network to deepen disappear at the same time avoid the gradient and the gradient of explosion,so as to solve the problem of generating high resolution CT images,the experimental results show that the network of the generated images and real image data distribution similarity is high enough,and on the visual effect is difficult to discern between true and false.Secondly,in terms of accurate image segmentation,this paper proposes a refined segmentation network for knee joint based on Mask_RCNN.Through two stages consisting feature region selection and image segmentation,the target is refined and segmented,and compared with the commonly used U-NET network,the results show that:With the same amount of training data,the convergence speed of the proposed method is 30%lower than that of U-NET,but the convergence process is more stable.The accuracy of femur and tibia segmentation of MASK_RCNN is%3 and%4 higher than that of U-NET,which reflects the advantages of this algorithm in small target detection.In addition,after data enhancement,the accuracy of MASK_RCNN for small scale target segmentation is improved by 1%,and the accuracy of U-NET is improved by 3%,which verifies the effectiveness of the data enhancement algorithm.Finally,to improve the accuracy and reduce the time of manual operation,a medical feature point identification algorithm based on multi-level point cloud network(PointNet++)was designed,and the network was applied to regression task to evaluate and analyze its effectiveness.The proposed network was compared with PointNet and 3D-UNET,the results showed that:The identification error of the network model designed in this study is 1mm lower than that of PointNet and 6mm lower than that of 3D-UNET.The variance of identification error is 0.7mm lower than that of 3D-UNET and 0.3mm lower than that of PointNet,showing stronger robustness.Identification time is less than 2s,much lower than manual operation time.In addition,automatic segmentation and manual segmentation were used for feature point identification respectively.The results show that the error of feature point identification using image segmentation and manual segmentation is similar,but the error variance of image segmentation is 0.1mm lower than that of manual segmentation,which verifies the feasibility of the image segmentation algorithm.The experimental results show that the automatic CT image processing method designed in this paper for navigation surgery of knee joint replacement can effectively improve the speed of preoperative image processing,reduce labor costs,and improve the degree of intelligence of surgical navigation system.