基于CT断层图像的脊椎骨分割方法研究

发布时间：2018-04-17 02:21

本文选题：癌症骨转移 + CT断层图像　；参考：《扬州大学》2017年硕士论文

【摘要】：癌症肿瘤在中晚期常常伴随着骨骼尤其是脊椎骨区域的转移。计算机辅助诊断系统能够依据CT断层图像提供的像素信息协助影像科医师及早发现脊椎骨转移的病灶。脊椎骨区域的分割作为CT断层图像处理中的重要步骤之一,一方面能够极大地减少脊椎骨区域配准以及骨转移识别的时间;另一方面其分割精度也直接影响着计算机辅助诊断系统的确诊率。论文针对多名骨转移患者CT断层图像,在研究数学形态学运算、连通区域标记算法以及图割算法理论的基础上,实现了脊椎骨区域的自动式、交互式以及两者相结合的分割,论文主要工作如下:(1)采用数学形态学运算对脊椎骨区域进行预处理,在保证骨转移患者的CT断层图像中各个器官、组织拥有明显灰度级差别的前提下,通过减弱图像二值化结果中的椒盐噪声,有效地避免了其它器官、组织的误分割,极大地提高了基于标记算法的脊椎骨区域分割的准确率。(2)提出一种基于标记算法的脊椎骨区域自动式分割方法,先后采用基于连通区域标记算法的二值化脊柱区域快速填补和基于行程标记算法的脊椎骨区域分割两个主要步骤,解决了由于癌症骨转移病灶的侵蚀致使患者CT断层图像中的脊椎体边缘模糊,无法保证高质量边缘信息提取的问题,使得计算机辅助诊断早期骨转移病灶成为可能。(3)基于经典Graph Cuts算法,提出了针对CT断层图像中脊椎骨区域分割的前景(目标)和背景区域初始种子点的选取方法。该方法在保证脊椎骨区域有效分割的条件下,选取矩形像素区域作为约束条件,避免了大范围的迭代运算,从而较大程度地降低了 Graph Cuts算法的处理时间,提高了分割的效率。(4)提出了一种改进Graph Cuts方法。该方法首先运用基于连通区域标记算法与数学形态学运算实现了脊柱区域分割,再利用Graph Cuts算法实现了脊椎骨区域的最终分割,解决了由于脊柱CT断层图像灰度级梯度幅值小导致经典Graph Cuts算法对脊椎体边缘分割不敏感的问题。(5)提出了一种基于区域像素点的真阳性率(TPR)和假阳性率(FPR)的分割结果评价方法。该评价法以日本产业医科大学多名资深影像科医师的人工分割作为参照标准,对基于经典Graph Cuts算法以及本文改进Graph Cuts方法的脊椎骨区域分割结果分别进行量化分析。分析结果表明:改进的Graph Cuts算法能够更好地分割脊椎骨的区域与边缘,为进一步的图像处理以及骨转移病灶的诊断提供了准确的病变组织信息。
[Abstract]:Cancer tumors are often associated with bone metastasis, especially in the vertebral region, in the middle and late stages.The computer aided diagnosis system can help the radiologist to detect the metastatic lesions of vertebral vertebrae early according to the pixel information provided by CT tomographic images.As one of the important steps in CT image processing, the segmentation of vertebral region can greatly reduce the time of registration and recognition of bone metastasis.On the other hand, its segmentation accuracy also directly affects the diagnosis rate of computer aided diagnosis system.In this paper, based on the research of mathematical morphology, connected area marking algorithm and image cutting algorithm, the automatic, interactive and combined segmentation of vertebral region is realized.The main work of this paper is as follows: (1) preprocessing the vertebral region by mathematical morphological operation to ensure that there are obvious grayscale differences in the organs and tissues of CT images of patients with bone metastasis.By reducing the salt and pepper noise in the binary image, the missegmentation of other organs and tissues is effectively avoided.It greatly improves the accuracy of vertebrae region segmentation based on marking algorithm. (2) an automatic segmentation method based on marking algorithm is proposed.There are two main steps: fast filling of spinal region based on connected region marking algorithm and segmenting of vertebrae region based on itinerary marker algorithm.It solves the problem that the edge of vertebral body in CT tomography image is blurred due to the erosion of cancer bone metastases, which can not guarantee the high quality edge information extraction.Based on the classical Graph Cuts algorithm, this paper proposes a method to select the foreground (target) and initial seed points of the background region in CT tomography.Under the condition that the vertebral region can be effectively segmented, the rectangular pixel region is chosen as the constraint condition, which avoids the iterative operation in a wide range, thus greatly reducing the processing time of the Graph Cuts algorithm.An improved Graph Cuts method is proposed.In this method, the spinal region is segmented based on the connected region marking algorithm and the mathematical morphology operation, and the final segmentation of the vertebra region is realized by using the Graph Cuts algorithm.This paper solves the problem that the classical Graph Cuts algorithm is insensitive to spinal body edge segmentation due to the small grayscale gradient amplitude of spine CT image. A method for evaluating the segmentation results of true positive rate and false positive rate based on regional pixels is proposed.The evaluation method is based on the artificial segmentation of many senior imaging physicians in Japan University of Technology and Medical Sciences as a reference standard. The results of the segmenting of the vertebrae region based on the classical Graph Cuts algorithm and the improved Graph Cuts method in this paper are quantitatively analyzed respectively.The results show that the improved Graph Cuts algorithm can better segment the region and edge of vertebrae and provide accurate tissue information for further image processing and diagnosis of bone metastases.
【学位授予单位】：扬州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R816.8;TP391.41

【参考文献】