Research On CT Image Segmentation Algorithm Based On Concatenated Convolutional Neural Network

Computed tomography images are also called CT images.CT image imaging technology has the advantages of clear imaging and fast imaging speed,and has become a common method in clinical examination.In today's life,80% of adults have spinal disease.The premise and important step of preoperative vertebral simulation is to segment a single vertebra from the spin CT images and mark it.However,the complex structure of the spine,the interlaced spatial positions,and the similarity of the adjacent vertebrae make it difficult to segment a single vertebra from the CT sequence image of the spine using conventional segmentation methods.Therefore,rapid and accurate segmentation of a vertebrae in computer-assisted surgery is an urgent and important work.The paper proposes a vertebral CT image segmentation algorithm based on concatenated convolutional neural network for automatic segmentation of spine CT images.Firstly,the paper studied the CT image preprocessing method,including CLAHE image enhancement,threshold segmentation and expansion operation,which can improve the quality of the input image and reduce the number of input voxel points.Then,the coarsely segmentation mixed residual connections full convolutional neural network(FCN)that roughly segment the vertebrae and other surrounding tissues to obtain the preliminary segmentation results of the vertebrae.Finally,the cascade fined segmented convolutional neural network(CNN)use to refine the coarse segmentation results and marks the different vertebrae to complete the automatic multi-vertebrae segmentation.By experimenting on the public dataset and the hospital dataset,the Dice similarity coefficient(DSC)of the CT image of the method proposed in this paper is 95%,and the accuracy of segmentation result classification can reach more than 98%.Compared with the advanced V-Net network,the segmentation algorithm proposed in this paper increases the segmentation accuracy by 5% and the segmentation time by about 2 minutes.The method can realize fast and accurate segmentation of the CT image of the spine.