Research On The Method Of Vertebrae Segmentation Based On CT Tomography

Terminal cancers are often accompanied by the metastasis to skeletons especially vertebrae.Fortunately,Computer Aided Diagnosis(CAD)system can assist radiologists to detect the lesions of early bone metastases based on CT images.As one of the important steps in CAD system,the segmentation of vertebrae not only reduce processing time of vertebral registration and bone metastasis diagnosis,but also directly affect the diagnostic accuracy of lesions of bone metastases.In this thesis,automatic and interactive segmentation methods,based on mathematical morphology,connected component labeling and Graph Cuts algorithms,are proposed to realize the segmentation of vertebrae in Computed Tomography(CT)images of patients who suffer from bone metastases.The main work is listed as follow.(1)Mathematical morphology operations are applied to do image preprocessing before performing vertebral segmentation.It can not only make sure obvious difference of gray level between spine and other organs,but also greatly remove the pepper noise in binary images.That surely avoids segmenting other organs and guarantees the accuracy of vertebral segmentation with following labeling algorithms.(2)An automatic method,which includes the region filling of vertebrae with connected component labeling algorithm and vertebral segmentation with run-length labeling algorithm,is proposed aimed to get vertebral region and borders with higher accuracy.Some traditional algorithms cannot segment the vertebrae with abundant information about its boundary because the surface of spine is destroyed by terminal cancer.The method effectively solves the problem,and make detecting bone metastases with CAD system possible.(3)A method to choose the seed points from foreground and background is firstly proposed based on traditional Graph Cuts algorithm.Under the condition that vertebrae can be segmented,it can greatly cut down the processing time of Graph Cuts algorithm with some pixels in rectangles.(4)An improved Graph Cuts is proposed to get vertebral segmentation efficiently and precisely.It applies labeling algorithm and mathematical morphology to realize spinal segmentation,and then uses Graph Cuts algorithm to attain the segmentation of vertebrae.It can solve the problem that traditional Graph Cuts algorithm is not sensitive to the boundary segmentation because of the gravitational gradient of spine in CT images of patients with bone metastasis.(5)On the basis of reliability,accuracy and statistical characteristics of region,an evaluation method considering pixels in region is proposed.In the evaluation,manual segmentation by experienced radiologists from University of Occupational and Environment Health is viewed as the standard,and segmentation results by traditional and improved Graph Cuts algorithms are quantitatively analyzed.As shown in evaluation,the vertebral segmentation with the improved Graph Cuts has the regions and boundaries with higher accuracy and efficiency.That can benefit the following image processes and provide detailed information about pathological tissue for radiologists.