CT Image Rib Segmentation Method Based On Convolutional Neural Network

With the development of computer technology and medical imaging technology,computer-aided diagnosis and treatment can reduce the workload of doctors and assist doctors in diagnosis.Human ribs include the whole abdomen and chest,with relatively stable shape.Rib segmentation is widely used in medicine.It can detect whether there is bone fracture or other abnormalities.Segmented ribs can also be used as reference objects to help locate other structures,such as liver,heart,etc.It can also provide reference for some quantitative tasks,such as lung volume estimation.The ribs are close to the scapula,spine,and sternum,and their gray values are similar.The traditional segmentation method cannot handle the segmentation of the bone junction well.Deep learning can automatically extract image features,and has made many achievements in the field of medical image segmentation,so deep learning is applied to rib segmentation.The main research content is divided into three parts:1)image preprocessing,combined with the characteristics that the rib in CT data consists of high pixel value surrounding low pixel value,laplace operator is used to enhance the image,so that the edge of rib in CT image is clearer;in addition,median filter is used to denoise CT image.2)Network model design,using the Unet network model and the 3DUnet network model to segment the ribs,compare the experimental results,and then improve the Unet network model.combined with the characteristics that different size and shape of the cross section of the ribs in CT data,pyramid pooling module is added to the Unet network,and an end-to-end segmentation model integrating multi-scale features is designed,and residual connection is added to the Unet model,so as to speed up the convergence of the network and avoid the problem of gradient disappearance.3)post-processing,the test results of different preprocessing image training networks are fused,and impurities are removed.The experimental results are evaluated quantitatively by Dice coefficient and Jaccard index,and the effect of difficult rib segmentation is especially analyzed.The results show that the improved network model is better than the basic Unet model,and the effect of combining three networks is better than that of a single network.