参数化的粘性形态学分水岭理论方法及其在医学肿瘤分割中的应用研究

Morphological watershed transform is an efficient way to segment image. Especially, it can produce closed and precise object contours. However, it often results in over-segmentation when it is directly applied to an original image. Usually, over-segmentation can be avoided in some degree by using pre-filtering, whereas the position of the segmented region contours may be offset. For the purpose of segmenting image without over-segmentation and region contours position bias, parameterized viscous morphological watershed segmentation is proposed, in which water is replaced by viscous fluid to perform watershed transform. It is well known that the viscousity of viscous fluid is determined by the flooding temperature. When temperature is higher, viscous fluid can fill the small irregulated relief, while at a lower temperature, it can only enter the larger relief. Such principle inspire us that one can adopt higher temperature fluid to flood the objects relief while lower temperature fluid to flood small image details and noise. By this way, irregulated local minimum can be eliminated and the relief minimum correspoding to object are preserved. However, how to control the viscousity of viscous fluid and what image feature parameter can be severved to control the viscousity are the key problems, In order to solve these problems, the model flooding an image relief with viscous fluid should be firstly constructed, and then select the image feature parameter to control viscousity. Simultaneously,the selected parameter should distinguish the image objects and irregulated details. Secondly, create the quantity relationship between the parameter and fluid viscousity, establish the parameter controled viscousity model. On the basis of the model, the function relationship between parameter and the size of structure element will be established, and such viscous structure element is used to simulate the process of viscous fluid flooding the image relief by morphological operators. This is to say that different relief position modification by morphological operator will adopt different sized structure element according to the current relief parameter. In this light, most local minimums caused by irregular details and noise will be removed, and region contour positions are not or slightly changed due to the object relief less or no modification. Finally, standard watershed transform is employed to implement segmentation on the basis of modified image relief. . On the basis of the proposed theory and method, we will develop tumour segementation software and integrate it into the hydronium computer-aided histopathology system. The aim is to precisely segment tumors from CT image, avoid losing the tumor fragment while protecting the healthy tissue from injury.

：形态学分水岭分割能够产生封闭且定位准确的图像目标区域轮廓，适合于医学肿瘤的分割需求，但存在过分割现象，对原始图像进行平滑滤波可以减轻或消除过分割，但容易使分割后的区域轮廓发生位置偏移。为此项目拟采用粘性流体替代经典分水岭中的水进行分水岭变换，通过粘性液体淹没图像地貌的流溢特性，建立图像地貌特征参数（高度级、体积等）与粘性流体温度以及粘性度之间的关系。从而使得对于目标区域采用温度较高的低粘度液体进行淹没，而对于非规则细节和噪声则采用温度较低的高粘性度液体进行淹没。在此模型的基础上，通过大小受特征参数控制的可变粘性结构元素对图像进行形态学运算，模拟并实现粘性液体淹没图像地貌的过程，一方面消除过分割，另一方面，实现目标区域轮廓的准确定位。.在参数化粘性形态学分水岭分割理论方法的基础上，将其应用于重离子放疗计划中肿瘤区域的分割，防止因分割定位不准确，造成对患者正常组织的损伤或肿瘤的遗漏。

在重离子放疗计划过程中，需要最大限度地将放射线剂量集中于患者肿瘤区域，而尽可能的减少对肿瘤周围正常组织器官的损伤。如果分割区域小于实际肿瘤，残留肿瘤得不到照射，反之，则会造成患者正常组织损伤。因此，肿瘤区域分割的准确与否，直接关系到患者的治疗效果。.项目利用形态学分水岭分割能够产生封闭且定位准确的图像目标轮廓的特点，对医学肿瘤进行分割定位。但由于形态学分水岭分割存在较严重的过分割，直接对医学肿瘤图像进行分割，无法分割出肿瘤区域。为此，项目在分析了分水岭产生过分割原因的基础上，提出了参数化的粘性形态学分水岭理论方法，并将其应用于医学肿瘤图像的分割。.项目的主要研究内容和结果如下：.（1）在分析粘性流体淹没图像地貌的基础上，建立了粘性流体淹没图像地貌的模型，构造了粘性流体温度控制的参数化模型；（2）通过理论分析和仿真实验，优选了合适的温度控制特征参数，并用利用选择的温度控制参数，建立了粘性流体温度的控制模型；确定了图像梯度（或灰度）与结构元素大小之间的关系相当于温度与粘性流体之间的对应关系，从而使粘性流体淹没图像地貌的过程转化为依据梯度不同而选择不同的结构元素大小进行形态学运算，定量描述了粘性流体淹没图像地貌的形态学运算过程；（3）定量建立了形态学圆形结构元素大小与温度控制参数之间的定量关系模型和计算模型；（4） 参数化粘性形态学分水岭运算模型建立与实现：建立了结构元素大小与图像特征参数之间的运算模型；建立了参数化可变结构元素对分解平面进行修正的形态学运算；（5）通过梯度图像分层，在每一层梯度级平面上利用参数化的结构元素进行相应的形态学闭运算，然后将修正后的图像平面进行重建，实现图像的参数化粘性形态学分水岭分割；（6）将提出的参数化粘性形态学分水岭理论方法应用于医学肿瘤影像的分割，保证肿瘤准确分割定位的前提下，提高重离子放疗的效果。.参数化的粘性形态学分水岭分割过程中，由于采用了大小受特征参数控制的可变粘性结构元素对图像进行形态学运算，一方面消除了造成过分割的区域局部极小值和噪声，减轻了过分割，另一方面，实现了目标区域轮廓的准确定位。.项目提出的分割方法已应用于重离子放疗计划软件系统中，能够在保证肿瘤准确定位的前提下分割出肿瘤目标。同时，该方法也可应用于对图像目标定位要求较高的图像处理场合。