知识引导几何动态轮廓线算法与MR心脏序列图像鲁棒分割

Knowledge Guided Geometric Active Contour Model and Robust Segmentation of Left Ventricle from Cardiac MR Image Sequences

从MR心脏三维动态序列图像中快速精确分割左心室内边界是心功能计算机辅助诊断的重要步骤。由于心室边界的模糊性,传统的基于灰度或曲线演化的方法很难保证分割结果的鲁棒和精确。在分割模型中整合解剖结构和医生经验的先验知识,对提高分割结果对噪声和模糊边界的鲁棒性,改善计算效率非常重要。本研究提出了一种广义模糊几何动态轮廓线分割算法(GF-GACM),并利用基于水平集的概率形状模型,整合医生手动分割训练集的先验知识。对多套临床数据集的实验结果显示,本研究算法的分割结果和专家手动分割结果比较在临床诊断允许误差范围内。

Segmentation of left ventricle from cardiac MR image sequences is the most important initial step in the computer aided diagnosis of heart function. Due to the fuzziness of ventricle borders, traditional gray level based or curve evolving methods often fail to ensure the robustness and accuracy of the results. It is commonly recognized that the incorporation of prior anatomy knowledge and expertise is very important to improve the robustness to noises and computational efficiency of segmentation algorithms. This paper proposes a new Generalized Fuzzy Geometric Active Contour Model by a level set based method of incorporating prior shape information from expert manual segmentation. Experiments on several clinical cardiac MRI data sets showed that results of the proposed algorithm are acceptable for clinical diagnosis.