基于改进Demons算法的三维肺部医学影像配准研究*

为实现相同个体在不同呼吸状态下产生较大形变的三维肺部医学影像配准，提出一种基于改进Demons算法的精确有效配准方案。首先，对待配准影像进行全局非刚性配准。通过尺度不变特征变算法对影像进行特征点提取与匹配，根据匹配结果计算变换参数，完成全局配准；其次，利用改进Demons算法对全局配准后的影像进行局部非刚性配准。使用改进的方案实现了人体肺部影像配准，并且肺部整体轮廓以及内部组织的配准结果较理想。配准前，影像间的均方误差值为25835.3，经配准后影像间均方误差值降为3726.31，均方误差值下降率为85.58%。提出的方案能够有效配准三维肺部影像，为对肺部呼吸运动估计以及呼吸功能分析提供良好的基础。

Three-dimensional Lung Medical Image Registration Based on Improved Demons Algorithm

To register three-dimensional pulmonary medical images of the same individual whose lung has large deformation under different respiration state, the paper puts forward an accurate and effective registration method based on improved Demons algorithm. Firstly, images were registered globally and non-rigidly. Feature points were extracted and matched by scale invariant feature transform algorithm. The global registration was finished according to the transformation parameter computed based on matching results. Afterwards, images after global registration were registered locally and non-rigidly utilizing improved Demons algorithm. Image registration of human lung was realized employing the improved method, and the registration results of the overall profile and internal organization of the lung were ideal. The mean-square error between images before registration was 25835.3 and it was reduced to 3726.31 after registration. The descent rate of mean-square was 85.58%. The proposed method effectively registers three-dimensional pulmonary images, which provides a solid foundation for doctors to estimate pulmonary respiratory movement and analyze respiratory function.