交互式GACV模型的肿瘤图像分割

目的:提取医学图像中肿瘤区域,用以测量肿瘤体积问题。方法:提出一种基于GACV(Geodesic-Aided C-Vmethod)的交互式模型。该模型首先人工选取感兴趣区域,并在区域内设定初始水平集与肿瘤内部种子点,然后在感兴趣区域上应用将图像梯度边缘信息与图像区域灰度特性统一到同一分割中的GACV模型,得到肿瘤的粗分割结果。最后为去除目标内外孔洞,提出一种无损边缘的膨胀搜索算法,作为细分割。结果:将该模型应用于不同形状的肿瘤图像中,能成功检测肿瘤轮廓。通过实验与其它活动轮廓分割方法结果对比,结果显示该模型在准确分割肿瘤边界与分割算法耗时方面均具有良好表现。结论:本文提出的分割方法能高效率、准确识别肿瘤区域。     

基于单水平集函数的细胞分割算法

目的：由于细胞图像十分复杂,传统的基于像素或者边界的图像分割方法难以精确的实现细胞分割。因此,需要设计一种可以实现细胞图像精确分割的方法。方法：结合大津分割算法和主动轮廓模型的优点,设计出一种基于单水平集函数的细胞分割算法,首先对细胞图像大津分割,其结果作为水平集函数的初始值,然后使用迭代法对水平集函数演化。采用MATLAB对显微镜下获取的细胞图像进行试验,将本文改进后的算法与常规的算法进行了对比。结果：与传统的水平集分割算法相比,本文方法对细胞图像分割结果更加准确,迭代次数减少一半左右,因此分割时间也减少了一半左右。结论：结合细胞图像的结构特点,利用大津分割结果作为主动轮廓模型的初始值,可有效解决主动轮廓模型因为初始值设置不当导致的分割缺陷问题,水平集函数能够跟踪拓扑结构变化,具有计算精度高、算法稳定、优化边界清晰光滑等优点,在本文中得到了充分的应用。因此本文所提出的算法能够高效地实现细胞图像的分割。     

基于改进的CV-RSF模型的甲状腺结节超声图像自适应分割算法

目的甲状腺结节超声图像的精确分割对甲状腺结节的良恶性诊断尤为重要。目前,对于甲状腺结节超声图像的分割,有学者提出利用主动轮廓模型分割算法,但是由于活动轮廓分割算法需要手动设置迭代次数,未实现模型的自适应性。因此,本文提出了一种基于改进的无边缘主动轮廓-局部区域可控的拟合(Chan-Vese-region scalable fitting,CV-RSF)模型的甲状腺结节超声图像自适应分割算法。方法选取南京同仁医院12例患者的甲状腺结节超声图像用于实验。首先,在无边缘主动轮廓(Chan-Vese,CV)模型中,引入一个基于梯度的边缘引导函数,根据面积变化率,自适应地获取甲状腺结节的粗分割轮廓;然后,将粗分割轮廓作为局部区域可控的拟合(region-scalable fitting,RSF)模型的初始轮廓,并根据面积变化率,自适应地获取甲状腺结节最终分割结果。将改进模型分割的结果与CV模型、RSF模型分割的结果进行比较,并分析甲状腺结节边缘清晰度对分割结果的影响。结果本文模型算法分割结果的平均迭代次数、平均面积重叠率、平均Hausdorff分别达到了134、90.34%、9.77,均优于CV模型、RSF模型的分割算法。结论该算法有效地分割出边缘清晰和不清晰的甲状腺结节超声图像,并解决手动设置迭代次数的问题,从而实现甲状腺结节的有效、准确、自动分割。     

一种新的舌癌图像快速自动分割方法

目的:研究一种新的舌癌图像自动分割算法以实现对舌癌肿瘤的快速准确分割。方法:通过引入一种基于局部均方差的自适应尺度算子实现演化曲线在演化过程中的自动调整,从而更高效率地向真实目标边界运动,并且克服舌癌肿瘤图像中目标边界不清和图像灰度不均匀等不良因素带来的影响。此外,为加快曲线的收敛速度,本文提出了一种新的能量项评估演化曲线轮廓内部和轮廓外部区域灰度的分布差异,以此引导曲线自适应地调整演化速度,减少完成分割任务所需的迭代次数。结果:使用本方法对22幅舌癌肿瘤MRI图像进行分割,分割结果与真实结果之间的重叠率Dice值为0.82,豪斯多夫距离HD值为1.732 mm。结论:将本文算法与其它现有的几种活动轮廓模型进行定性和定量对比分析,实验结果表明本文算法在对细节及弱边缘灰度的处理上表现更加优异,可用于舌癌肿瘤的精确分割,为临床分析提供辅助信息。     

基于层次上下文活动轮廓的三维CT肝脏图像分割

本文提出一种新的活动轮廓算法,即层次的上下文活动轮廓(HCAC),并将其应用于三维CT肝脏全自动分割。HCAC是一种基于机器学习的算法,可以分为两个阶段:第一阶段,即学习阶段,给定一套腹部三维CT训练图像以及对应的手动肝脏分割结果,利用上下文特征将每次的自动分割结果向手动参考分割结果映射,迭代学习得到一组纠错分类器;第二阶段,即分割阶段,首先将待分割图像用基本的活动轮廓进行分割,分割结果输入第一个纠错分类器,输出第一个形状模型,然后结合图像信息和当前形状模型,采用上下文活动轮廓(CAC)进行再一次分割,得到的分割结果输入第二个纠错分类器,输出第二个形状模型,结合图像信息和当前形状模型,再次采用CAC进行分割。如此迭代分割,随着形状模型的逐步精确,最终我们可以取得准确的肝脏分割。本文实验结果表明,随着迭代的深入,我们取得越来越好的分割结果。在三维CT肝脏图像分割中,我们通过6次迭代,就可以取得较好的肝脏分割。     

基于光学相干层析的视网膜图像分割

利用光学相干层析（optical coherence tomography,OCT）技术可以得到清晰的视网膜层状结构,实现视网膜层状结构自动分割功能是解决OCT技术应用于视网膜疾病诊断的一项基础问题。本文通过图像平滑、峰值探测、Snake模型、贪婪算法和样条插值等综合技术,对OCT视网膜图像进行分割,自动提取层状结构轮廓并获取视网膜厚度分布图。将以上算法应用于24例正常人眼底图像,并与专家手动标记轮廓提取的厚度相比,结果证实上述视网膜自动测量算法与专家人工标记取得较好一致性。本文提出的测量算法有望应用于视网膜变异性评估。     

基于活动轮廓模型的图像分割算法综述

活动轮廓模型是一种重要的图像分割技术,它利用底层信息,并结合高层先验知识,实现对复杂目标轮廓的自动分割。自Kass等提出该思想以来的20多年中,活动轮廓模型在理论研究和应用方面均取得长足发展。首先,介绍活动轮廓模型的发展历程,重点阐述并分析典型的参数活动轮廓模型和几何活动轮廓模型,进而扼要介绍混合活动轮廓模型和快速求解算法;随后,从理论基础、分割效果、算法效率以及应用等方面,比较两类模型之间的区别与联系;最后,对活动轮廓模型未来的发展趋势进行展望。     

基于交叉熵和GVF-Snake的子宫肌瘤高强度聚焦超声图像自动分割算法

目的:高强度聚焦超声(HIFU)广泛应用于肿瘤无创治疗,目前手术中多用超声成像技术进行导航,但由于HIFU图像对比度低,信噪比低以及目标边界模糊等缺点,HIFU图像的目标识别与分割是重点也是难点,所以需要提出能够自动快速获得HIFU图像肿瘤轮廓的分割方法。方法:GVF-Snake模型算法能够有效地利用超声图像的局部与整体信息实现边界的准确定位,非常适用于HIFU图像分割,但是作为参数活动轮廓模型,GVF-Snake对初始轮廓的依赖性较强,而且通常采用手画初始轮廓,增加了人为因素对试验结果的干预。针对GVF-Snake的相关特性,本文提出用二维最小交叉熵阈值分割法来提取初始轮廓。交叉熵是度量两个统计概率分布之间信息量差异的物理量,分别表征分割前后图像中像素特征向量的概率分布,当原始图像和分割图像之间的信息量差异最小时,便得到最优阈值。使用二维最小交叉熵算法求得初始轮廓后,进而使用GVF-Snake模型收敛,得到最终结果。结果:该算法对HIFU图像中子宫肌瘤的识别与分割具有较为理想的效果,统计结果显示灵敏度平均值达到87.56%,标准化的Hausdorff距离指数平均值达到4.95%,整体算法的运行时间平均值达到2.16 s。结论:该分割算法通过GVF-Snake自动生成初始轮廓,避免了人为干预,整体分割算法快速精准,取得了较好的实验结果,为其在HIFU设备的应用奠定了基础。     

基于对称区域生长和边缘梯度的视神经纤维的分割

在视神经横切面图像中,将每个神经纤维的内外边界进行精确分割是视神经形态分析的重要环节,提出一种基于对称区域生长和髓鞘边缘梯度的有效分割算法.该算法分两步进行,首先根据交互方式下选取的种子点,由对称区域生长算法实现轴突分割,然后在轴突轮廓模型基础上,髓鞘外轮廓在髓鞘平均边缘梯度引导下进行演化,实现自动分割.与K-均值聚类,局部阈值和水平集等其他算法的实验结果相对照显示,该算法分割获得的轴突和髓鞘轮廓与实际轮廓相吻合,其分割结果可以作为后续神经纤维形态分析的基础.     

基于区域自适应形变模型的CT图像牙齿结构测量方法研究

针对三维计算机断层扫描(CT)体数据的牙齿分割问题,本文提出了一种基于区域自适应形变模型的CT图像牙齿结构测量方法。本文方法结合了自动阈值分割、CV活动轮廓模型和图割方法,利用自动阈值分割实现牙冠的分割与定位,然后利用牙冠分割结果作为初始轮廓逐层分割牙齿。在分割难度最大的牙根上采用CV活动轮廓和图割互补的方法实现了牙根的准确分割。实验结果表明本文提出的牙齿结构测量方法能够准确地自动分割出牙齿的牙冠部分,进而在牙冠分割基础上快速准确地分割出牙颈和牙根。本文提出的牙齿结构测量方法能够准确地从临床CT牙齿数据中分割提取牙齿结构,鲁棒性强、精度高,可以有效辅助医生的临床治疗。     

Cell image segmentation with kernel-based dynamic clustering and an ellipsoidal cell shape model

In this paper, we propose a novel approach to cell image segmentation under severe noise conditions by combining kernel-based dynamic clustering and a genetic algorithm. Our method incorporates a priori knowledge about cell shape. That is, an elliptical cell contour model is introduced to describe the boundary of the cell. Our method consists of the following components: (1) obtain the gradient image; (2) use the gradient image to obtain points which possibly belong to cell boundaries; (3) adjust the parameters of the elliptical cell boundary model to match the cell contour using a genetic algorithm. The method is tested on images of noisy human thyroid and small intestine cells.     

Segmentation of carotid artery in ultrasound images: method development and evaluation technique

Segmentation of carotid artery lumen in two-dimensional and three-dimensional ultrasonography is an important step in computerized evaluation of arterial disease severity and in finding vulnerable atherosclerotic plaques susceptible to rupture causing stroke. Because of the complexity of anatomical structures, noise as well as the requirement of accurate segmentation, interactions are necessary between observers and the computer segmentation process. In this paper a segmentation process is described based on the deformable model method with only one seed point to guide the initialization of the deformable model for each lumen cross section. With one seed, the initial contour of the deformable model is generated using the entropy map of the original image and mathematical morphology operations. The deformable model is driven to fit the lumen contour by an internal force and an external force that are calculated, respectively, with geometrical properties of deformed contour and with the image gray level features. The evaluation methodology using distance-based and area-based metrics is introduced in this paper. A contour probability distribution (CPD) method for calculating distance-based metrics is introduced. The CPD is obtained by generating contours of the lumen using a set of possible seed locations. The mean contour can be compared to a manual outlined contour to provide accuracy metrics. The variance computed from the CPD can provide metrics of local and global variability. These metrics provide a complete performance evaluation of an interactive segmentation algorithm and a means for comparing different algorithm settings.     

结合区域信息的分段活动轮廓模型

提出了一种结合区域信息的分段活动轮廓模型，利用边缘信息迅速找到对象的大体轮廓，然后结合区域统计信息使模型精确收敛到对象边缘。分段的层次化变形有效的利用了图像的全局和局部信息，使用仿射变换使模型的局部以同一种变换方式变形，提高模型对噪声和伪边缘的鲁棒性，同时保持模型轮廓形状的一致性。在精确匹配阶段利用区域统计信息重新定义模型的外部能量，采用自适应的搜索区域确定方法，提高了算法的效率和进入凹边缘的能力。试验表明本模型运算速度快，抗噪声和避免陷入局部极小值的能力较强，有较好的分割效果。     

伽玛刀治疗计划系统中基于链码的图像轮廓提取算法

在伽玛刀治疗计划系统中，为了给三维重建提供直观的参考依据，要求准确地对体表、病灶和其他组织进行轮廓提取。针对目前该系统中以手工轮廓提取为主的现状，本文提出一种基于链码的轮廓提取算法对其进行比较改进，并给出断层扫描图像应用本算法的结果。结果表明本算法的使用提高了治疗计划系统中轮廓提取的自动化程度，并能满足精度要求。     

用改进的Snake算法检测血管内外轮廓

动态轮廓模型 (Snake算法 )是一种较好的目标轮廓检测方法。但是传统的Snake算法在对其能量函数进行优化时 ,不能检测多目标图像 ,并且不能检测凹形目标和内轮廓。本文首先对图像进行自动Snake初始化 ,然后在Snake的成长过程中加入向心力因子。实验结果表明 ,新的算法能够检测多目标轮廓和凹形目标及内轮廓 ,优于传统的Snake算法。     

Prostate segmentation algorithm using dyadic wavelet transform and discrete dynamic contour

Knowing the location and the volume of the prostate is important for ultrasound-guided prostate brachytherapy, a commonly used prostate cancer treatment method. The prostate boundary must be segmented before a dose plan can be obtained. However, manual segmentation is arduous and time consuming. This paper introduces a semi-automatic segmentation algorithm based on the dyadic wavelet transform (DWT) and the discrete dynamic contour (DDC). A spline interpolation method is used to determine the initial contour based on four user-defined initial points. The DDC model then refines the initial contour based on the approximate coefficients and the wavelet coefficients generated using the DWT. The DDC model is executed under two settings. The coefficients used in these two settings are derived using smoothing functions with different sizes. A selection rule is used to choose the best contour based on the contours produced in these two settings. The accuracy of the final contour produced by the proposed algorithm is evaluated by comparing it with the manual contour outlined by an expert observer. A total of 114 2D TRUS images taken for six different patients scheduled for brachytherapy were segmented using the proposed algorithm. The average difference between the contour segmented using the proposed algorithm and the manually outlined contour is less than 3 pixels.     

基于裂隙扫描图像的角膜表面三维重建

角膜是人眼的重要组成部分，眼外科中对角膜表面检查一般采用基于Placido盘的角膜地形图法和裂隙扫描角膜地形图法，但这两种方法的结果都未能获得直观的角膜表面三维模型。本研究设计的一种基于裂隙扫描图像的角膜表面三维重建算法，实现了对角膜前后表面的重建，并能快速地计算出角膜前后表面的几何参数以及转换成传统的等高线图和地形图，方便了医生的诊断。该算法首先采集角膜的裂隙扫描图像序列，然后在图像集中提取出每层角膜的轮廓，进而采用轮廓线重构算法得到角膜表面模型。在对裂隙图像提取闭合轮廓线时，根据角膜区域的颜色差异，采用了色彩分割的算法；针对角膜轮廓线是凹的特点，对轮廓线重构算法做了改进，以便能构造出正确的角膜表面三角形网格。     

Accuracy and sensitivity of finite element model-based deformable registration of the prostate

PURPOSE: Evaluate the accuracy and the sensitivity to contour variation and model size of a finite element model-based deformable registration algorithm for the prostate. METHODS AND MATERIALS: Two magnetic resonance images (MRIs) were obtained for 21 prostate patients with three implanted markers. A single observer contoured the prostate and markers and performed blinded recontouring of the first MRI. A biomechanical-model based deformable registration algorithm, MORFEUS, was applied to each dataset pair, deforming the second image (B) to the first image (A). The residual error was calculated by comparing the center of mass (COM) of the markers with the predicted COM. Sensitivity to contour variation was calculated by deforming B to the repeat contour of A (A2). The sensitivity to the model size was calculated by reducing the number of nodes (B', A', A2') and repeating the analysis. RESULTS: The average residual error of the registration for B to A and B to A2 was 0.22 cm (SD: 0.08 cm) and 0.24 cm (SD: 0.09 cm), respectively. The average residual error of the registration of B' to A' and B' to A2' was 0.22 cm (SD: 0.10 cm) and 0.25 cm (SD: 0.10 cm), respectively. The average time to run MORFEUS on the standard and reduced model was 3606 s (SD: 7788 s) and 56 s (SD: 16 s), respectively. CONCLUSIONS: The accuracy of the algorithm, equal to the image voxel size, is not affected by intraobserver contour variability or model size. Reducing the model size significantly increases algorithm efficiency.     

An automated segmentation method for three-dimensional carotid ultrasound images

We have developed an automated segmentation method for three-dimensional vascular ultrasound images. The method consists of two steps: an automated initial contour identification, followed by application of a geometrically deformable model (GDM). The formation of the initial contours requires the input of a single seed point by the user, and was shown to be insensitive to the placement of the seed within a structure. The GDM minimizes contour energy, providing a smoothed final result. It requires only three simple parameters, all with easily selectable values. The algorithm is fast, performing segmentation on a 336 x 352 x 200 volume in 25 s when running on a 100 MHz 9500 Power Macintosh prototype. The segmentation algorithm was tested on stenosed vessel phantoms with known geometry, and the segmentation of the cross-sectional areas was found to be within 3% of the true area. The algorithm was also applied to two sets of patient carotid images, one acquired with a mechanical scanner and the other with a freehand scanning system, with good results on both.