基于耦合可变模型的膀胱壁分割方法
发布时间：2019-01-03 19:33
【摘要】：在膀胱磁共振(Magnetic Resonance, MR)图像中，膀胱壁的准确分割对临床应用和医学研究具有重要意义。膀胱癌是一种发病率和复发率都很高的疾病，早期的检测非常重要。膀胱内外壁之间的厚度的异常增加，可以作为膀胱肿瘤检测的一个重要指标，而衡量膀胱壁厚度的一个最基本且最重要的工作是对内外壁进行准确地分割。目前在临床应用中，通常是医务人员手动地对其分割。但这是一项繁重而且耗时的工作，尤其是随着医学影像技术的发展，需要处理的图像越来越多，只依靠手动分割是不可能完成的。因此，本文研究课题为高精度的计算机自动膀胱壁分割方法。 本文针对MR图像的膀胱壁分割的挑战，逐步研究并提出有效的解决方法。首先，针对膀胱MR图像中存在的伪影问题，利用梯度的方向信息提出基于方向性梯度的水平集模型用以区分膀胱内壁和伪影边缘，在一定程度上减少了伪影对内壁分割的影响；针对复杂的外部组织，利用膀胱壁的区域信息构建出耦合水平集模型同时分割内外壁，利用较准确的内壁分割修正外壁分割；并且在耦合水平集模型中加入最小壁厚度的先验知识，防止内外零水平集的重叠或交叉。然后，针对靠近膀胱顶部或底部的层中存在的部分弱边界的问题，本文提出利用上一层的分割结果作为形状先验，并自适应地约束本层的分割，，初步解决了水平集在弱边界的泄漏问题。在验证了形状先验对于膀胱壁分割的有效性之后，进一步提出了更为准确的形状先验构建方法，即部分稀疏形状模型，利用部分可靠的轮廓构建出完整可靠的形状先验；并且提出了扇区驱动的水平集模型，更为全面的考虑了不同区域和不同演化阶段对约束力的需求。最后，将本文所提出的部分稀疏形状模型拓展到经典的主动形状模型(Active Shape Model, ASM)中，解决了由于部分弱边界造成的错误搜索的问题，证明了该模型的普适性与有效性。 本文的主要创新点：1)提出耦合方向性水平集模型；2)提出自适应形状约束的水平集模型；3)提出部分稀疏形状约束的扇区驱动的水平集模型；4)将部分稀疏形状模型推广到ASM中。我们的方法在15组不同病人的共167层的数据上进行实验，膀胱壁的分割精度达到：内壁的P2C值为1.06±0.28mm，DSC值为0.98±0.01，外壁的P2C值为1.46±0.42mm，DSC值为0.97±0.01，与现有方法对比，证明了本文所提出的方法的有效性与准确性。
[Abstract]:In (Magnetic Resonance, MR) images of bladder, accurate segmentation of bladder wall is very important for clinical application and medical research. Bladder cancer is a disease with high incidence and recurrence rate. Early detection is very important. The abnormal increase of the thickness between the inner and outer walls of the bladder can be regarded as an important index for the detection of bladder tumor. The most basic and important work to measure the thickness of the bladder wall is to segment the inner and outer wall accurately. At present, in the clinical application, it is usually the medical personnel to divide it manually. But this is a heavy and time-consuming task, especially with the development of medical image technology, more and more images need to be processed. Therefore, a high-precision automatic bladder wall segmentation method is studied in this paper. Aiming at the challenge of bladder wall segmentation in MR image, this paper studies and proposes an effective solution step by step. Firstly, aiming at the artifact problem in bladder MR image, a level set model based on directional gradient is proposed to distinguish bladder inner wall from artifact edge, which reduces the influence of artifact on inner wall segmentation to a certain extent. For the complex external tissue, the coupled level set model was constructed to segment the inner and outer wall simultaneously, and the inner wall was used to segment the outer wall. A priori knowledge of minimum wall thickness is added to the coupled level set model to prevent the overlap or crossover of the internal and external zero level sets. Then, aiming at the problem of partial weak boundary in the layer near the top or bottom of the bladder, this paper proposes to use the segmentation result of the upper layer as a shape priori, and adaptively constrains the segmentation of the layer. The leakage problem of the level set at the weak boundary is solved preliminarily. After validating the validity of shape priori for bladder wall segmentation, a more accurate shape priori construction method, that is, partial sparse shape model, is proposed, and a complete and reliable shape priori is constructed by using partially reliable contour. A sector-driven level set model is proposed, which takes into account the binding requirements of different regions and different evolution stages. Finally, the partially sparse shape model proposed in this paper is extended to the classical active shape model (Active Shape Model, ASM), which solves the problem of error search caused by partial weak boundary, and proves the universality and validity of the model. The main innovations of this paper are as follows: 1) the coupling directional level set model is proposed; 2) the level set model with adaptive shape constraints is proposed; 3) the sector driven level set model with partial sparse shape constraints is proposed; 4) the partial sparse shape model is extended to ASM. Our method was tested on 167 layers of data from 15 different groups of patients. The division accuracy of bladder wall was 1.06 卤0.28 mm DSC 0.98 卤0.01 and 1.46 卤0.42 mm respectively, and that of internal wall was 1.06 卤0.28 mm DSC value was 0.98 卤0.01, and that of outer wall was 1.46 卤0.42 mm. The DSC value is 0. 97 卤0. 01, which proves the validity and accuracy of the proposed method.
【学位授予单位】：中国科学院研究生院（西安光学精密机械研究所）
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R445.2;R737.14
本文编号：2399785
[Abstract]:In (Magnetic Resonance, MR) images of bladder, accurate segmentation of bladder wall is very important for clinical application and medical research. Bladder cancer is a disease with high incidence and recurrence rate. Early detection is very important. The abnormal increase of the thickness between the inner and outer walls of the bladder can be regarded as an important index for the detection of bladder tumor. The most basic and important work to measure the thickness of the bladder wall is to segment the inner and outer wall accurately. At present, in the clinical application, it is usually the medical personnel to divide it manually. But this is a heavy and time-consuming task, especially with the development of medical image technology, more and more images need to be processed. Therefore, a high-precision automatic bladder wall segmentation method is studied in this paper. Aiming at the challenge of bladder wall segmentation in MR image, this paper studies and proposes an effective solution step by step. Firstly, aiming at the artifact problem in bladder MR image, a level set model based on directional gradient is proposed to distinguish bladder inner wall from artifact edge, which reduces the influence of artifact on inner wall segmentation to a certain extent. For the complex external tissue, the coupled level set model was constructed to segment the inner and outer wall simultaneously, and the inner wall was used to segment the outer wall. A priori knowledge of minimum wall thickness is added to the coupled level set model to prevent the overlap or crossover of the internal and external zero level sets. Then, aiming at the problem of partial weak boundary in the layer near the top or bottom of the bladder, this paper proposes to use the segmentation result of the upper layer as a shape priori, and adaptively constrains the segmentation of the layer. The leakage problem of the level set at the weak boundary is solved preliminarily. After validating the validity of shape priori for bladder wall segmentation, a more accurate shape priori construction method, that is, partial sparse shape model, is proposed, and a complete and reliable shape priori is constructed by using partially reliable contour. A sector-driven level set model is proposed, which takes into account the binding requirements of different regions and different evolution stages. Finally, the partially sparse shape model proposed in this paper is extended to the classical active shape model (Active Shape Model, ASM), which solves the problem of error search caused by partial weak boundary, and proves the universality and validity of the model. The main innovations of this paper are as follows: 1) the coupling directional level set model is proposed; 2) the level set model with adaptive shape constraints is proposed; 3) the sector driven level set model with partial sparse shape constraints is proposed; 4) the partial sparse shape model is extended to ASM. Our method was tested on 167 layers of data from 15 different groups of patients. The division accuracy of bladder wall was 1.06 卤0.28 mm DSC 0.98 卤0.01 and 1.46 卤0.42 mm respectively, and that of internal wall was 1.06 卤0.28 mm DSC value was 0.98 卤0.01, and that of outer wall was 1.46 卤0.42 mm. The DSC value is 0. 97 卤0. 01, which proves the validity and accuracy of the proposed method.
【学位授予单位】：中国科学院研究生院（西安光学精密机械研究所）
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R445.2;R737.14
【参考文献】
相关期刊论文 前1条
1 段侪杰;田珍;梁正荣;包尚联;袁克虹;;磁共振虚拟膀胱镜中膀胱壁分割与壁厚估算[J];清华大学学报(自然科学版);2010年09期
本文编号：2399785
本文链接：https://www.wllwen.com/yixuelunwen/fangshe/2399785.html
