【作者】 周健；

【导师】 罗立民；

【作者基本信息】 东南大学 ， 生物医学工程， 2005， 硕士

【摘要】 本文重点研究了放疗感兴趣区域的分割和可视化。为了减轻医生手动勾画感兴趣区域的繁重劳动,提出了半自动和自动分割两种解决方法。半自动方法:手动合并分水岭分割结果中的过分割块,继而标识出感兴趣区域。在分水岭算法的预处理阶段,比较了多种滤波和梯度生成算法,确定用强滤波器滤波,用形态学多尺度梯度算法生成梯度图。接下来,实现了一种快速的基于链码的分水岭算法;改进了传统的基于浸没模型的分水岭算法,在运算时间几乎没有影响的情况下,大大抑制了过分割。针对分割结果中依然存在过分割,比较了3种合并准则,得出使用灰度、面积准则的结论。自动方法: SOM和上下文相关两种神经网络相结合以分割图像,然后使用若干判定准则,标识出腹部感兴趣区域。为了增强效果,介绍了放疗感兴趣区域的可视化。运用轮廓线表面重建方法构建出三维感兴趣区域,帮助医生直观地观察人体内部病灶及组织器官的形态、位置和尺寸。实现了CT模拟技术的两个重要应用:数字重建射线图像和射野方向观视。为了增强BEV的效果,提出了通过图像融合,把放疗感兴趣区域的三维图像融入到DRR中,得到了具有立体效果的BEV图。更多还原

【Abstract】 Region of Interest’s (ROI) segmentation and visualization are studied in this thesis. In order to set radiologists free from the tedious work of locating the ROI, we propose two methods, which segment objects semi-automatically or automatically. The first one: using watershed algorithm coupled with manually incorporation. Several filters and gradient formation methods have been considered during the pre-processing stage. In the meanwhile a fast watershed method based on chain code has been introduced and the traditional watershed method base on flood has been improved to suppress the over-segmentation greatly without increasing computing time. 3 incorporation rules have been compared to process the segmentation. The second one: using SOM and contextual neural networks to segment image and recognizing objects by judging rules. Besides, for the purpose of improve the ROI’s visual effect, visualization technology is also introduced. We reconstruct 3-D human body model using contours to help radiologists observing the appearance, position and size of the focus and organs, and implemented the Digitally Reconstructed Radiographs (DRR) and Beam’s eye view (BEV) which are important in CT simulations. To enhance the effect of BEV, we blend the 3-D image of the ROI into DDR using image fusion to get the 3-D BEV image.更多还原