中文摘要

生物医学成像的研究方兴未艾，成像技术层出不穷，为生物组织的影像学描述提供了新的工具。融合显示方法的研究有助于多模态信息的整合和新模态的运用，因此在临床实践和基础研究中有广泛的应用价值。融合显示方法的设计，需要结合具体应用的需要和源图像的特点。本论文针对生物和医学领域的三个应用，对灰度图像与伪彩图像的融合显示进行了研究。 生物相衬图与荧光图的融合有助于蛋白质亚细胞结构的定位和功能研究。融合图像应在真实反映细胞内蛋白质分布的基础上，保留足够的亚细胞结构的定位信息，着力改善荧光背景对融合图像明度和清晰度的影响。 研究结合...>> 详细

生物医学成像的研究方兴未艾，成像技术层出不穷，为生物组织的影像学描述提供了新的工具。融合显示方法的研究有助于多模态信息的整合和新模态的运用，因此在临床实践和基础研究中有广泛的应用价值。融合显示方法的设计，需要结合具体应用的需要和源图像的特点。本论文针对生物和医学领域的三个应用，对灰度图像与伪彩图像的融合显示进行了研究。 生物相衬图与荧光图的融合有助于蛋白质亚细胞结构的定位和功能研究。融合图像应在真实反映细胞内蛋白质分布的基础上，保留足够的亚细胞结构的定位信息，着力改善荧光背景对融合图像明度和清晰度的影响。 研究结合广义亮度-色相-饱和度(Generalized Intensity-hue-saturation，GIHS)快速算法、非降采样轮廓波变换(Nonsubsampled Contourlet Transform，NSCT)和透明度的概念，提出了GIHS框架下变权重多尺度融合的流程。融合方法的设计在引入了NSCT理论的基础上，结合绝对值最大的高频系数融合准则，提出了以双伽马函数为原型的低频系数融合准则。 实验部分结合Otsu分割算法和视觉信患保真度(Visual Information Fidelity， VIF)指标，设计了分区域的相似度量化实验。通过117组拟南芥相衬图和荧光图的融合显示实验，研究了多尺度变换对图像融合显示效果的影响。实验结果表明：非降采样技术和二维滤波器设计在一定程度上提高了图像融合的效果，但是多尺度变换的方向性与融合效果的提升关系不大。在该方法和多种传统方法融合显示效果的比较实验中，量化结果与融合实例中的观察结果吻合，都表明了该方法的有效性和相比传统方法的优越性。 脑部磁共振成像(Magnetic Resonance Imaging，MRI)与单光子发射计算机断层成像(Single Photon Emission Computed Tomography，SPECT)的融合实现了人体结构和功能信息的整合，尤其有助于SPECT图的解读和运用。 融合图像应在真实反映脑部功能变化的基础上，保留足够的结构定位信息，着力于改善SPECT图黑色背景对融合图像明度和清晰度的影响。研究将图像的融合显示归结为线性优化问题，运用单纯形法得到权重函数，用于源图像亮度分量低频系数和高频系数融合准则的设计。研究还设计了算法的可交互特性，提出了全局透明度和全局高频透明度，用以实现源图像之间的渐变以及对融合图像细节表现的控制。 实验探讨了基于高阶多项式近似的权重函数的估计，相比基于双伽马函数的近似，其优势在于放宽了对权重函数形态的限制。在正常脑部29个层面MRI图像和SPECT图像的融合实验中，融合实例的观察结果与量化实验结果吻合，都表明了该方法的有效性和相比传统方法的优越性。继而以传统透明度技术为参照，验证了融合显示方法的可交互特性设计。两组临床实例证实了提出的融合显示方法的实用价值。最后结合实验，初步探讨了两种权重函数设计方法的优缺点和通用性。 B型超声图与超声弹性图的融合不仅有助于生物组织力学特性分布的定位、理解和运用，而且扩大了超声无损诊断的应用范围。融合图像应在保持源图像理解模式的基础上，提高融合图像对源图信息的表现力，尤其是对B型超声灰度图像中低频结构细节的表现力。 研究基于色貌模型理论，设计了均一明度的伪彩显示方法，并将色貌属性的预测用于图像的融合显示。在国际发光照明委员会CIECAM02色貌模型的运用中，通过假定已知明度、色相和饱和度，着重推导了图像的CIEXYZ色彩空间表示。 实验研究了计算机可显示的色彩集合在色彩空间和色貌模型中的分布。其结果表明：灰度图像的明度范围受融合图像中允许的色彩集合限制；色貌属性中的饱和度调节了融合结果的色相分辨率和明度范围。通过49组仿真图像的量化实验，验证了提出的基于CIECAM02色貌模型的融合显示方法的有效性。结合2组临床实例的融合结果，证实了该方法的应用价值，及其相比传统方法的优越性。最后对应用范围进行了扩展，初步探讨了其在脑部MRI图像融合中的可行性。 关键词：图像融合显示，广义亮度-色相-饱和度框架、非降采样轮廓波变换，变权重，伽马函数，单纯形法，色貌模型，生物医学图像

外文摘要

Researches on biomedical imaging have been unfolding with the advent of new modalities, which provides new tools for the description of biological tissues. Integrated visualization benefits the fusion of multimodalities and facilitates the interpretation of new modalities for a wide range of applications. It is specifically designed to meet the requirements by using the characteristics of original images. This dissertation focuses on the crucial problems in three biomedical applications, when ...>> 详细

Researches on biomedical imaging have been unfolding with the advent of new modalities, which provides new tools for the description of biological tissues. Integrated visualization benefits the fusion of multimodalities and facilitates the interpretation of new modalities for a wide range of applications. It is specifically designed to meet the requirements by using the characteristics of original images. This dissertation focuses on the crucial problems in three biomedical applications, when integrating the gray image and pseudo-color image. Combination of the fluorescent image and the phase contrast image is valuable for the localization and functional studies of the protein. The fused result aims to restrain the black background of the fluorescent image, and display a complete distribution of the targeted proteins with sufficient location information of subcelluar structures. In the view of transparency, a variable-weight method is proposed in the generalized intensity-hue-saturation frame. The intensity components of original images are first analyzed by the nonsubsampled contourlet transform (NSCT). Subsequently, the low-frequency coefficients are combined as the rule based on a piecewise function consisting in two Gamma functions, while the high-frequency coefficients are combined as the maximum selection rule. Using the Otsu method and the visual information fidelity (VIF), the region-based assessment of similarity is designed to reveal the similarity between the fused image and the original images. Validation experiments on 117 sets of Arabidopsis images are for two purposes: the impact of the multiscaled analysis in the fusion and the comparison among different methods of integrated visualization. It was demonstrated that the invariance and the design of true two-dimensional filters rather than the directionality attribute to the fused effect. Besides, both the quantitative results and the application examples demonstrated the superiority of the proposed method over the traditional ones. The single photon emission computed tomography (SPECT) image alone is difficult to understand and utilize, since anatomical structures are absent from the data. Integrated visualization attempts to locate functional information of the SPECT image by the structural information of the magnetic resonance (MR) image. The fused result aims to restrain the black background of SPECT, and display a complete distribution of functional variation with sufficient location information of anatomical structures. After approximating the weight by a polynomial function, integrated visualization can be regarded as a linear optimization problem in which the simplex method is applied to determine the coefficients of the weighting function. The rules are then proposed for the combination in the multiscaled spaces. Interactive approaches are designed for the gradual variation between original images with the global transparency, and the control of detail performance with the global high-frequency transparency. In the experiments, the weighting function was first estimated. Its shape had less restraints than that based on two Gamma functions proposed in the former application. The similarity assessment then evaluated several different methods on a normal brain atlas composed of 29 slices. The observations in the example matched the quantitative results, and demonstrated the superiority of the proposed method over the traditional methods. Two clips showed the interactive property of the proposed method, while two medical cases demonstrated its clinical values. Besides, two methods for designing the S-shaped weighting function were also compared, ie. the simplex method and the method based on two Gamma functions. Combined with the B-mode ultrasound image, the elastography image can be better understood and utilized, which also expands the application range of ultrasonic non-destructive examination. Except preserving the original styles of interpretation, the fused result also aims to improve the ability in expressing the original images, especially the low-frequency structures in the B-mode ultrasound image. The fused method is proposed based on the color prediction of the color appearance model. It additionally involves the pseudo-color display with uniform lightness. Here the formulae are exhaustively derived from the CIECAM02 color attributes to the CIEXYZ color channels, after the assumption of the known attributes including lightness, hue and saturation. Based on different color spaces and the color appearance models, the distribution of color aggregations was studied in the experiment. It was concluded that the range of lightness was determined by the available color aggregation for the fused image, and the saturation controlled the resolution of hue and the range of lightness. The similarity assessment was performed on 49 sets of simulated images. Its result was reasonable, and thus revealed the effectiveness of the proposed method. Two medical cases demonstrated the clinical values of the proposed method and its superiority over the traditional methods. Besides, its feasibility in fusing two high-resolution structural images was preliminarily approved based on the simulated MRI data. Keywords: integrated visualization, generalized intensity-hue-saturation (GIHS), nonsubsampled contourlet transform