一种基于三维约束的医学图像序列分割新方法

A New Medical Image Sequence Segmentation Method Based on 3D Constraints

基于三维图像具有三维平滑、连续变化等特性,提出一种新的医学图像序列分割方法。该方法在计算过程中,只需手动设置第一幅图中主动形变模型的关键点位置。在其他图像中,首先采用预测和视频运动估计中常用的块匹配法优化主动轮廓模型的初始化位置,然后采用偏三维约束和梯度矢量流(GVF),从初始化位置开始在图像中进一步迭代收敛得到每幅图像中的最终轮廓位置。在一般的二维平滑基础上,达到三维平滑、三维分割的作用。与光流法、一般预测法等方法的实验结果相比,本方法可以显著提高分割的准确性及速度。

A new medical image sequence segmentation algorithm was proposed based on the characteristics of 3D images, i.e. the surface is smooth and the deformation of the contour is continuous in 3D space. The algorithm just needed to set a few key points of active contour model in the first image. There were two steps for the method in each image. Firstly, we optimized the initial position by forecasting and block-matching which were often used for video motion estimation. Secondly, we got the final contour position after iterations from initial position by partial 3D constraints and gradient vector flow（ GVF）. Thus 3D reconstruction with 3D smoothness was gained. The results showed that the new algorithm could remarkably improve the accuracy of the segmentation and reduce computation time in comparison with the optical flow field and the normal forecast methods.