Medical Cervical X-ray Image Registration Based On Point Feature Description

With the development of society,people's working intensity is increasing,and the working time is gradually extended.As time passes,the increasing intensity brings many lesions of the human body,and the spine is the most vulnerable part of people.Cervical vertebra,as a bridge connecting the head and thoracic cage,which is characterized by high mobility and easy to strain and prone to lesions.Anterior cervical discectomy and fusion(ACDF)are often used in the treatment of cervical spondylosis.It is important to check the range of motion of the vertebral body regularly after surgery of determine the outcome of the operation.Existing tests such as CT scanning of cervical vertebra and cervical hyperextension flexion position of X ray film,etc.The former patients receiving radiation dose is larger increase risk of cancer and the price is expensive,although the radiation is small.But the anatomical mark(such as vertebral body Angle)on the display is not clear,than the upper and lower end plate cannot be sure of vertebral body and a lot of anatomical marks is discontinuous or irregular shape,which brought trouble to further determine the ACDF surgery effect.For this reason,we plan to use the limited image marker points to use the registration algorithm for image matching,so as to improve the measurement accuracy of vertebral movement,provide more effective information and help for doctors' diagnosis.Improving the accuracy of determining about the intervertebral bodies have been fused in the second visit after anterior cervical surgery.This paper's main contents are as follows:1.For the problem of different angles and at different times of the X-ray image data is hard to compare analysis of cervical spine,using the image point and line features and the traditional algorithm to gain a key description of the effective characteristics.With the aid of all kinds of the different image feature points unique description,the algorithm can complete the matching of feature point.By means of random sampling principle,the algorithm can screen out error matching feature points,and then it offered to the point mean square error as the evaluation standard,which is guided by line features.An iterative screening registration scheme was designed and implemented,which improved the stability of X-ray cervical image registration and provided effective information for doctors' diagnosis.2.In view of the peoblem about complexity of X-ray cervical vertebra images and poor visual effect,using the rich point feature information in the data and the deep learning registration method train the deformation field from the source image to the target image to represent the displacement of each point.But the displacement of each point has different contributions to the registration robustness.The spatial transformation network is used to fix the number of the parameters,and then the network is placed behind the deformation field?So the displacement of each point in the deformation field can be trained,and the more effective point information can get more weight for the training and registration of each parameter of the projection matrix.The basic idea is to train the affine transformation matrix or deformation field of the image by using the information of the feature points in the images.Using this feature of spatial transformation network,an unsupervised registration algorithm was designed.Based on the cervical image data provided by China-Japan Friendship Hospital,a method of extracting point features based on spatial transformation network was proposed for X-ray cervical image registration.The experimental results have better visual effects and provide more information for doctors' diagnosis.