Segmentation Of The Spine Region Of CT Image Based On 3D Edge Detection And 3D Region Growth Method

In recent years,bone metastasis has been recognized as one of the leading causes of death Bone metastasis is a malignant tumor located outside bone tissue.It destroys bone tissue through interaction with osteoblasts,osteoclasts and bone stromal cells.Bone metastasis carcinomas occur mostly in the spine.Early diagnosis is an important means for the treatment of vertebral metastasis.In clinical diagnosis,radiotherapy needs to determine the location and volume of the treatment target.Accurate segmentation of therapeutic targets can greatly reduce the time of radiotherapy and surgery,so it is necessary to accurately segment the vertebral region based on three-dimensional image processing technology.In this paper,we propose an image segmentation method based on three-dimensional edge detection and three-dimensional region growing techniques based on CT of multiple bone metastasis patients.A segmentation algorithm based on three-dimensional region growing quickly and accurately extracts the spinal region from the CT image.The ROI(region of interest)is divided into two parts,the spine and the spinal canal.The main contents of this paper are as follows:1.In order to eliminate noise and small gray scale difference in the spine region,unclear boundary and complex texture structure,efficient three-dimensional(3D)Gaussian smoothing in frequency domain convolution is used to smooth the image processing.Non-linear mapping is used to enhance the contrast of the region,effectively avoiding the influence of the darkness of CT images in some patients.Cutting the ROI area greatly improves the subsequent segmentation speed without destroying the image quality.2.A segmentation algorithm based on approximate 3D Canny edge detection to determine the initial ROI growth range is proposed.Firstly,the 3D Sobel operator is used to calculate the gradient.The approximate Canny method finds the edge by searching the local maximum of the gradient of the region;then,Using the histogram of the resulting global gradient,the initial seed point set is automatically selected and edge information is provided for subsequent region growth.3.For the spine segmentation,a method combining edge detection and region growth is proposed.First,iteratively updates the average gray level of the seed point set by selecting the set of initial seed points in the low gradient range and the high gray level pixel;then,based on the judgment of the 26 neighborhood gray value,the process of region growing begins with The voxel position of the small gradient magnitude ends at a position with a large gradient magnitude;finally,when no similar grayscale voxels are added to the set and the previously detected edge is reached,the growth stops.It solves the limitations of region-based segmentation and the possibility of under-segmentation and over-segmentation.4.Considering that the shape of the spinal canal is similar to a circle,a Hough transform algorithm is employed to locate the ROI to find the initial seed point.In the sagittal plane,the spinal canal region is divided into three segments,and the segmentation based on the 3D region growth algorithm solves the effect of large curvature of the spinal canal on the over-segmentation.5.A method for evaluating the segmentation results based on the target detection and evaluation ratio(IOU)of regional pixel points is proposed.The evaluation method uses the manual segmentation of several senior imaging physicians at the Japan University of the Industrial Medical as a reference standard to quantify the spine regions segmented based on the 3D edge detection and 3D region growing algorithms.The experimental results show that the algorithm guarantees the robustness of regional growth results and maintains a balance between segmentation accuracy and speed,providing clinical information for further diagnosis and treatment.