Research On Cervical Spine X-ray Image Segmentation Algorithm Based On Deep Learning

In recent years,as a common spinal disease,cervical spondylosis has shown a high incidence trend.The commonly used method for clinical diagnosis of cervical spondylosis is X-ray imaging.At present,for this type of images,most of the clinical indicators that need to be measured for the diagnosis of cervical spondylosis are realized manually.This depends heavily on the experience and technology of the operator,and there are problems such as high workload for doctors and low accuracy of indicators.Therefore,based on artificial intelligence,image processing and other related knowledge,the study of automatic measurement methods for clinical indicators of cervical X-ray images will help improve the accuracy and intelligence of cervical spine lesion diagnosis,help doctors reduce work intensity,and make it an important theory.significance and practical value.In view of this,this thesis,based on deep learning,deeply studies the key core algorithm in the automatic measurement technology of cervical clinical indicators-cervical spine segmentation algorithm for X-ray images.Based on the analysis of the convolutional neural network image segmentation method,the algorithm optimization is focused on how to improve the segmentation accuracy of cervical X-ray images,from the improved U-Net cervical image segmentation algorithm and the multi-scale Deep Labv3+ efficient segmentation based on prior knowledge The algorithm is studied in two aspects.The specific research content of this thesis is as follows:(1)Based on the improved U-Net cervical spine image segmentation algorithm.In order to improve the detailed information of the image,a method based on the combination of U-Net and VGG16 model is proposed to achieve accurate segmentation of the image.First,replace the encoding part with an improved VGG16 model,which adds a Dropout layer on the basis of removing the fully connected layer,so as to obtain deeper image information while reducing the number of parameters;secondly,in the adjacent A jump connection is added between the upper and lower layers,and the fine-grained and coarse-grained features are combined to enhance the detailed information of the image;finally,the CBAM module is introduced in the decoding part,and the weight of the image channel and space is learned to obtain accurate detail features.Through the inspection of clinical X-ray images,it is proved that the proposed method has better segmentation accuracy,thus verifying the feasibility of the new algorithm model.(2)Multi-scale Deeplabv3+ efficient segmentation algorithm based on prior knowledge.In order to improve the segmentation quality of cervical spine images,a segmentation method that applies multi-scale and prior information to the model is proposed.First,a multi-scale structure and an adaptive ECA module are introduced into the ASPP module,so that significant performance gains can be obtained while improving the utilization of image pixels;second,prior knowledge is added after the multi-scale ASPP high-efficiency module to help obtain contextual images information;finally,inspired by the pyramid idea,the Laplacian pyramid structure is introduced in the decoding part,and by extracting multi-level low-level features,more detailed information can be obtained to improve the segmentation accuracy of cervical X-ray images.Based on the experimental verification of clinical data and comparative analysis with other classic algorithms,it is proved that the new algorithm model has a good segmentation effect,and the feasibility and effectiveness of the proposed method are verified.