一种MR膀胱图像增强后基于图论的分割策略
Segmentation strategy for enhanced MR cystography based on graph theory

为了从具有低信噪比和运动伪影的虚拟膀胱镜核磁成像图像MR(magnetic resonance)中获得精确的膀胱分割结果, 该文提出一种针对连续采集的多套短时膀胱MR影像的处理策略。该策略选取一套图像作为配准基准, 对其他短时图像先后进行仿射变换和分层B样条变换配准, 最后对配准结果平均, 获得增强图像。继而利用基于闭集模型的图割方法分别对基准图像和增强图像进行分割。该文利用计算机生成的模拟图像以及临床MR图像测试该策略。实验结果表明: 模拟增强图像和MR增强图像的信噪比分别增大到原来的3.26倍和2.17倍, 图像信噪比在配准后明显提高; 对增强图像使用该文提供的分割方法获得了更好的分割结果。实验证明该文提出的策略能通过配准方法获得高信噪比、高对比度、较少伪影的膀胱增强影像。
Abstract：Fast magnetic resonance (MR) bladder scans with artifacts and low signal to noise ratios (SNR) are used to precisely segment and achieve the bladder wall. The short scans are registered to a selected reference using an affine transformation followed by a hierarchical B-spline registration. The average of the registration results is the motion-corrected image. The graph cut method based on a closed-set model is then used segment the bladder MR image. The strategy is evaluated using both computer-generated images and clinical MR images. The results show that the average motion-corrected image with a high SNR (i.e., 3.26 for the simulated images and 2.17 for the clinical images) and less artifacts followed by a graph-cut segmentation tends to have a more accurate result. This strategy reduces the artifacts and improves the SNR to provide high resolution segmentation of the bladder wall.