基于平均密度投影和平移高斯模型的肺结节检测与分割算法

针对 CT图像中肺结节与血管粘连导致分割困难的问题,提出了一种基于平均密度投影和平移高斯模型的肺结节检测与分割算法。首先通过对二维CT序列图像作平均密度投影(AIP),融合局部三维特征生成AIP图像,然后利用阈值分割和形态学方法对结节轮廓进行粗分割,最后通过建立平移高斯模型来拟合肺结节,从而实现对肺结节的精确分割。对30个血管粘连性肺结节CT图像的实验结果表明,本文算法与专业医师标记区域的面积交迭度达到91%,能够实现对粘连型肺结节的有效分割,但对于灰度较弱且体积较小的肺结节仍存在漏检的风险,需要后续进一步研究。     

基于多级小波分解的两类纹理图像分割技术

通过对纹理图像传统分割方法的分析,研究了农作物图像的边缘自动分割问题,提出了采用多级小波分解方法的两类纹理图像分割技术.解决了传统的基于Laws能量特征方法存在的分割边缘粗糙以及时间效率低等缺点.仿真结果表明,应用基于多级小波分解的纹理分割方法,能快速得到准确的边缘,为获取大规模农作物的机械导航路径提供了一种新的思路.     

一种基于形态学可变元的肺实质分割方法

能极大提高肺癌患者生存率的诊断方式是肺癌的早期诊断,基于当前肺癌的早期诊断都依赖于恶性肺结节的检测,而肺结节检测算法数据准备环节最重要的一步是肺实质分割的现状,本文提出一种基于形态学可变元的肺实质分割方法。传统的形态学肺实质分割方法在分割大多数正常肺部CT时,能够达到优良的效果,但是在面对肺部情况复杂,CT层位于肺部顶端或底端时,传统的形态学肺实质分割方法便会失效,此时便需要人工分割。本文提出的这一新方法,能极大提高肺实质分割准确率,摆脱需要人工分割的局面。     

面向肺癌CAD的CT图像疑似病灶检测算法

CT图像中疑似结节病灶区域的分割和提取是肺癌CAD系统的关键和难点.本文提出一种疑似结节病灶自动检测算法,首先对原始CT图像进行有效、准确的肺实质分割,根据肺结节、气管和血管具有不同的几何特征,构造了一组不同尺度的类圆形结构元素,采用多尺度形态学滤波方法对ROI进行初始分割,再根据各ROI的大小构造相应尺度的二维高斯模板,对各ROI区域进行自适应局部高斯模板匹配,以进一步剔除假阳性.实验结果表明,该算法可以有效地提取出CT图像中类圆形的疑似结节病灶,具有较高的灵敏度和较低的漏诊率,可以为医生诊断早期肺癌病灶提供辅助信息.     

有复杂背景的条码图像实时处理的研究

将图像分割技术应用于条码自动识别上,首先进行预处理,运用图像腐蚀的方法,使图像上的文字等复杂背景变成一些孤立点,再通过中值滤波器滤除这些孤立点.此时图像中条码的相对位置不变,用模板匹配的方法对图像进行区域查找,在原图像上可分割出条码区域.由于采用模板匹配的方法,存在计算量大的问题,而DSP的运算功能强,采用DSP实现图像处理,可以达到实时处理条码图像,实现条码图像的自动识别.     

一种彩色铁谱图像自动分割方法

铁谱图像有效分割是实现铁谱图像自动化分析的基础,但彩色磨粒图像背景颜色相对单一,磨粒区域同时包含亮区域和暗区域且与背景色差较大,且部分磨粒存在黏连的情况,因此铁谱图像的有效分割难以实现。根据彩色铁谱图像特点,提出基于两次K-means颜色聚类分离磨粒区域与背景区域后,再对磨粒区域采用改进的分水岭算法分割黏连磨粒的图像分割方法。该方法解决了铁谱图像中亮暗磨粒同时存在的情况下磨粒提取不完全的问题,并实现了黏连磨粒的分割。实验结果验证了该方法的有效性。     

一种结合域自适应的图像语义分割算法

语义分割为图像分配像素级稠密标签,对场景理解具有重要作用,是视觉研究核心任务之一,涉及自动驾驶、医学成像等实际应用.现有基于深度神经网络的图像语义分割方法的训练需要大量标记数据,这些数据的收集和标记成本高昂,这很大程度上限制了此类方法的实际应用.为解决此问题,这里使用成本较低的计算机生成并标记的逼真的合成数据训练深度神经网络.但真实图像与合成图像在分布域上存在的差异会降低模型性能,因此这里使用一种对抗学习方法来实现域的自适应,以解决上述问题.又因为语义分割的结构化输出描述了源域与目标域的空间相似性,所以这里选择在语义分割的输出空间上实现域自适应方法.在此基础上,这里又于不同级别的空间构建多级域自适应网络以提升模型性能.     

喷墨印花纹理图像的期望最大化聚类分割算法

针对聚类分割算法对喷墨印花纹理图像存在的局限性,提出了一种结合期望最大化(EM)的喷墨印花纹理图像聚类分割算法(CSA).首先,将空间相关性引入聚类分割中,利用自回归模型表征纹理同质区域;然后,为了提高分割模型参数估计的精度,将分块标定机制引入期望最大化算法中,实现了参数极大似然估计的迭代算法,解决了不完全数据参数估计问题;最后,利用数据集分块并进行聚类,使同类元素具有较高的相似度,从而对图像中的像素进行了归类划分,并将得到的结果进行了合并,实现了目标图像的正确分割.实验结果表明,和传统的聚类分割算法相比,该算法能更好地解决喷墨印花纹理图像的分割问题.     

基于形态学重建和GMM的球团颗粒图像分割

机器视觉技术的发展为颗粒粒径的自动测量提供了一种有效方法,但是,重叠颗粒的图像分割问题仍有待进一步解决。针对这一问题,提出一种基于形态学重建和高斯混合模型的球团颗粒图像分割算法。首先利用似圆度将单独颗粒和重叠颗粒进行区分;根据重叠颗粒图像距离变换特征建立了高斯混合模型;为实现无监督的聚类,采用形态学重建结合聚类有效性指标的方法获得最佳聚类数目,并利用期望极大(EM)算法进行求解;最后采用圆拟合的方法对缺失的球团颗粒轮廓进行重构,实现了对重叠球团颗粒的分割。实验结果表明,该算法能够有效地对重叠颗粒进行分割,分割正确率评价指标AC为93.6%,明显优于现有的对比算法,为基于机器视觉的球团颗粒粒径分布测量奠定了基础。     

基于活跃度的脉冲耦合神经网络图像分割

针对脉冲耦合神经网络(PCNN)在图像分割中需多次人工调整网络参数的问题,提出了一种基于PCNN模型的图像自动分割方法.分析了图像中影响PCNN参数设置的因素,提出了一种图像自适应分块策略.将图像划分为内部复杂程度相近的子块,克服了同一参数无法同时对图像中复杂度差异较大的不同区域准确分割的不足.利用本文提出的局部图像活跃度(ADLA)指标对不同子块自适应地确定PCNN模型参数,有效解决了传统PCNN图像分割时需要人工确定关键参数的问题.最后,采用最大二维Tallis熵准则从分割后的二值结果序列中选择最佳结果.实验表明,本文方法的分割结果轮廓清晰、完整,即使在对比度低、背景呈大范围内连续变化等复杂情况下,也具有优异的性能.与传统PCNN分割方法相比,本文方法能自动、快速、准确地确定PCNN模型参数,且区域一致性测度(UM),区域对比度(CR),形状测度(SM),综合指标(CI)等客观评价的量化指标均优于传统PCNN分割方法12％以上.     

Cell therapy for fibrotic interstitial pulmonary disease: experimental study

Parte superior do formulário Digite um texto ou endereço de um site ou traduza um documento. The aim of this study is to evaluate the histological changes in lung parenchyma of pigs affected by interstitial lung disease induced after the infusion of bone marrow mononuclear cells (BMMCs). Ten female swines were submitted to pulmonary fibrosis induced by a single dose of intratracheal bleomicine sulfate. Animals were arranged into two groups: Group 1: induced‐disease control and Group 2: cell therapy using BMMCs. Both groups were clinically evaluated for 180 days. High‐resolution computed tomography (HRCT) was performed at 90 and 180 days. BMMC sampling was performed in cell therapy group at 90 days. Euthanasia was performed, and samples were collected for histology and immunohistochemistry. The 90‐days HRCT demonstrated typical interstitial lesions in pulmonary parenchyma similarly to human disease. The 180‐days HRCT in Group 1 demonstrated advanced stages of the disease when compared with Group 2. Immunohistochemistry analysis suggests the presence of pre‐existent vessels and neoformed vessels as well as predominant young cells in the injured parenchyma of Group 2. Immunohistochemistry analysis suggests that cell therapy would promote a reconstructive response. Histology and HRCT analysis suggest a positive application of swine as a model for a bleomicine inducing of fibrotic interstitial pulmonary disease. Microsc. Res. Tech., 2011. © 2011 Wiley‐Liss, Inc.     

Ultrastructural characteristics of the lung of <i>Melanophryniscus stelzneri stelzneri</i> (Weyenberg, 1875) (Anura,Bufonidae)

The lung of the toad, Melanophryniscus stelzneri stelzneri was studied using scanning and transmission electron microscopy. In M.s.stelzneri the parenchyma forms a polygonal network arrangement, therefore the parenchyma is edicular. These spaces are delimited by the interconnection of third order septa which are covered by respiratory epithelium. Small patches of ciliated epithelium without goblet cells appear irregularly distributed on the septa. The respiratory epithelium consists of one type of pneumocyte, which shows characteristics of both type I and type II alveolar cells of higher vertebrates. The pneumocytes are irregular in shape and possess attenuated cytoplasmic processes, which spread around the capillaries to form the outer layer of the air-blood barrier. These cells contain different types of cytoplasmic bodies: electron dense bodies, multivesicular bodies and lamellar bodies. Dense bodies are probably the precursors of lamellar bodies and the multivesicular bodies are incorporated into the latter. Neuroepithelial bodies appear randomly distributed over the septa. These bodies are separated from the lumen of the lung by thin cytoplasmic processes of neighbouring pneumocytes. The air-blood barrier consists of three layers: epithelium, interstitial space and endothelium.
The relatively simple pulmonary structure of M.s.stelzneri is due to a lower degree of partitioning of the pulmonary lumen in comparison to the lung of other bufonid anurans, could be correlated with a well developed cutaneous and buccopharingeal respiration. The testing of this hypothesis awaits further studies.     

结合多尺度圆形滤波与MS聚类 的疑似结节分割

肺部CT图像中疑似病灶感兴趣区域(ROI)的准确分割是肺部计算机辅助检测/诊断(CAD)的重要环节。本文提出结合Hessian矩阵滤波的均值漂移聚类肺部疑似病灶ROI区域分割算法。对原图像进行多尺度Hessian矩阵圆形滤波,图像中圆形的疑似结节病灶区域得到滤波增强、直线形的气管/血管区域得到抑制,将Hessian矩阵滤波后的形状特征、灰度、空间位置3种信息引入特征空间,将均值漂移聚类的核函数分解为3种特征信息所分别对应的核函数乘积形式,最后采用自适应计算带宽的方法确定每个待分割疑似区域的带宽进行均值漂移聚类分割。对来自LIDC等127个包含不同类型肺结节的病例进行实验,实验结果表明引入Hessian矩阵圆形滤波信息的均值漂移聚类能够分割出与血管或气管相连或者交叉的结节区域,去除ROI中包含的非结节区域,能有效分割出基于灰度信息难以分割的毛玻璃型(GGO)结节;对于3种类型的结节区域:血管相连结节(VPN)、毛玻璃型结节(GG0)、孤立性结节(SPN)分割平均准确率分别为92.80%、86.13%、95.08%。     

应用电子束CT低剂量扫描在冠状动脉CT成像病人中筛查早期肺癌

目的:探讨新一代电子束CT(e-SPEED)在行冠状动脉CT成像检查病人中联合筛查早期肺癌的扫描方法及应用价值。方法:对104例健康查体者先行肺部低剂量电子束CT扫描,扫描参数为130kV,895mA,50ms,层厚6mm,层距3mm,重建层厚3mm。平扫后再经静脉注射对比剂进行CT冠状动脉成像检查,其中48例(心率≥70次/分)行电子束CT扫描,扫描参数为Multiphase Dual方式,140kV,1010mA,100ms,层厚1.5mm;57例(心率<70次/分行16排螺旋CT扫描,扫描参数为120kV,370mA,500ms,层厚0.625mm。对比剂选用非离子型Ommipaque 350mgI/mL。肺平扫图像采用肺算法及标准算法重建,由两位放射诊断医师在工作站显示器上进行观察,统计病变检出例数并观察肺叶、段支气管显示情况;并对冠状动脉CT增强图像以VR、MIP、CINE、CTVE方式重建后进行观察,利用其薄层扫描的优势重点观察肺门结构及支气管开口情况,并观察可疑病变的强化情况。所有图像存贮至图像存储与传输系统(PACS)。结果:100%肺叶、段支气管可显示,共检出3mm以上非钙化肺结节4例,其中1例经活检证实为I期周围型肺癌。结论:在冠状动脉CT成像检查同时配合电子束CT低剂量平扫,适于联合筛查早期肺癌,并有助于提高肺内非钙化小结节及支气管细微病变的检出。     

Preventive approach against drug-induced pulmonary fibrosis through the suppression of epithelial-mesenchymal transition

A number of drugs induce pulmonary injury and subsequently lead to serious lung diseases such as pulmonary fibrosis as the adverse drug reactions. However, an effective preventive approach against drug-induced pulmonary fibrosis has not been established due to poor understanding of common preventive targets in a variety of drugs showing pulmonary toxicity. Epithelial-mesenchymal transition (EMT), a cellular phenotypic change of the epithelial to mesenchymal state, contributes to the development of pulmonary fibrosis through the conversion of damaged alveolar epithelium into myofibroblasts. As several drugs with pulmonary toxicity have been reported to induce EMT, EMT serves as a bridge between the drugs and pulmonary fibrosis. Accumulated evidence supports the potential of EMT as a preventive target against drug-induced pulmonary fibrosis. Additionally, since there are mechanistic differences between the main pharmacological effect and EMT induced by the drug, prevention based on EMT suppression would be possible and would contribute to continuous clinical treatment with the drug to avoid EMT-mediated serious pulmonary fibrosis. Furthermore, targeting EMT seems to be adequate for exerting a preventive effect since EMT in damaged alveolar epithelial cells occurs prior to the development of the pathophysiological state of the whole lung in a bleomycin-induced lung injury rat model. This viewpoint deals with the benefits and perspectives of preventive approaches against druginduced pulmonary fibrosis through the suppression of EMT, which has rarely been addressed.     

Evaluation of lanthanide tracer methods in the study of mammalian pulmonary parenchyma and cardiac muscle by electron energy-loss spectroscopy

Lanthanum (La) has widely been used as a tracer to study the integrity of plasma membranes. With conventional transmission electron microscopy (cTEM), the absence of electron scattering deposits from the cytoplasm has generally been assumed to reflect an intact cell membrane. However, the application of electron spectroscopic imaging (ESI) and electron energy-loss spectroscopy (EELS) reveals that electron scattering deposits may be present which do not contain La. However, La could be detected in regions of pulmonary parenchyma and cardiac muscle that were devoid of electron scattering deposits. Therefore, to exclude misinterpretations based on cTEM the application of microanalytical techniques is strongly recommended for the study of the integrity of plasma membranes by means of La tracers. In addition, ESI and EELS are shown to distinguish between different tracers in simultaneous applications of La and terbium (Tb) which were used at the different faces of the pulmonary air-blood barrier. The analysis of the distribution of both tracers which form electron scattering deposits, indistinguishable by cTEM, may help us to understand the different functional significances of cellular alterations of both cellular borders of the barrier. As was shown for La, however, strictly controlled conditions are mandatory during the fixation procedure because an increase in the incubation time to more than 1 h in samples of pulmonary parenchyma may result in the occurrence of La deposits within the cytoplasm. In the absence of electron scattering deposits, the presence of La in glycogen granules and ribosome-containing areas of various types of alveolar septal cells even after 15 min incubation indicates that the absence of deposits does not necessarily correspond to the absence of the tracer.     

A new method for pulmonary edema research: Scanning electron microscopy of frozen-hydrated edematous lung

We present an integrative method for studying alveolar fluid accumulation during pulmonary edema using scanning electron microscopy (SEM) of frozen-hydrated lung. We induced edema in dogs by fluid over-loading the vascular system while monitoring vascular and airway pressures. Terminally, we freeze-fixed the lungs in the open thorax while controlling airway pressure. Using a cryosample preparation chamber attached to a scanning electron microscope equipped with a low temperature stage, we in vacuo freeze-fractured the lung, gold coated the fractured surface and used secondary electron emission scanning electron microscopy to examine the frozen-hydrated lung with scanning electron micrographs of freeze-dried and airway fixed, critical point dried lung. We discuss the advantages of frozen-hydrated scanning lung preservation for pulmonary edema research, especially in regard to quantifying alveolar fluid distribution by stereo photogrammetry, and suggest areas of future research to improve the method.     

Electron micropscopic appearances of rapidly frozen lung.

Analysis of pulmonary structure-function relationships by microscopy requires that the lung be fixed under carefully controlled physiological conditions, since lung structure is extremely responsive to the relationship between airway and vascular pressures. Unfortunately, standard post-mortem fixation techniques leave some doubt as to the exact relationship between these pressures during fixation. This problem can be circumvented by stabilizing lung structure by rapid freezing under carefully controlled physiologic conditions. Using ethylene glycol in a freeze substitution technique we have developed procedures which yield a degree of preservation compatible with the high degree of resolution of the electron microscope. These can be used to obtain a more detailed understanding of pulmonary structure-function relationships under well-defined physiological conditions.     

Computer-assisted stereology: point fraction of lung parenchyma and alveolar surface density in fetal and newborn sheep

The aim of this study was to determine the accuracy of a new computer-assisted stereological technique in obtaining structural information of the lung. We compared the point fraction of lung parenchyma (Pp) and alveolar surface density (Sv) obtained by established manual point/intercept counting methods and compared them with those obtained using a computer-assisted method. Lung tissues obtained from normally grown fetal sheep (n = 6) and from newborn lambs with severe lung hypoplasia (n = 5) were inflation fixed via the trachea and processed for light microscopy. In verification-of-technique experiments, Pp and Sv correlated well with known values. There was a significant linear correlation between manual and computer-assisted stereological measurements for values of Pp (r2 = 0.92) and Sv (r2 = 0.98). Our data lead us to believe that the computer-assisted stereological technique described in this study provides accurate estimates of Pp and Sv and hence may be a valuable tool for evaluating the effects of factors upon structural development of the lung.