Research On Lumbar Detection And Tracking Based On Learning

With the continuous development of digital image technology and computer vision technology,the use of computer technology to process and analyze medical image data to assist doctors to diagnose the disease has been gradually popularized.As lumbar instability is a common and serious health problem,a new method to help diagnose lumbar spine instability is proposed in this paper.The method uses DVF image sequence of lumbar spine to process and analyze and the main research work is as follows:In view of the problem that the initial state of the lumbar spine tracking method needs to be manually labeled,a lumbar detection method based on convolution neural network is proposed.In this method,the original DVF image is filtered and denoised,and the resolution of DVF image is increased.In the off-line training phase,a large number of lumbar samples are used to train the neural network classifier.During the detection,the use of the Hough transform in the binary DVF images will help find the corner of the lumbar spine to get the angle parameters,and then the lumbar anatomical statistical data is used to obtain the initial candidate detection area.Then the initial candidate detection area is fed into the convolutional neural network classifier to obtain the detection results.In the experiment,the method performs high lumbar detection accuracy,so it can applies to situations that need high precision.In view of the poor robustness performance of current algorithms in the tracking for lumbar,a kind of on-line update lumbar tracking method based on stacked auto-encoder is proposed.The method can be used to express the deep characteristics of the general objects in the off-line training.During the online tracking framed by particle filter,tracking results of current frame will be obtained by the auto-encoder,then the weights of neural network parameters are updated online.The method effectively improves the robustness of the algorithm and reduces the possibility of tracking drift.It performs a strong ability of lumbar recognition.The techniques above can be applied to the DVF image with low contrast and moreambiguous situations to detect part of the lumbar.Moreover,it displays a strong recognition ability to detect lumbar instability symptoms accurately in the process of tracking,which can be used as a clinical diagnosis of lumbar instability in a very effective assistant.