Abstract: Segmentation of cardiac MRI images is the fundation of computer-aided diagnose and analysis of cardiac function.The key to segment cardiac MRI images accurately is to delineate the contour of left ventricle correctly.This paper proposes a novel method to delineat the contour of left ventricle.Firstly，the cardiac MRI images are smoothed using an adaptive smoothing algorithm，which smoothes the images without shifting the image edge information.Then，the images is clustered with k-mean clustering to find homogeneous region.At last，the contour of left ventricle is delineated by geometric active contour based on variation level set methods.Experimatal results demonstrate that the method is robust to the noise，and the interference of surrounding tissues of left ventricle and the results is accurate.

Key words: medical image segmentation, cardiac Magnetic Resonance Imaging（MRI） image, adaptive smoothing, K-mean cluster, active contour model, variational level set method

摘要： 心脏磁共振图像的分割是心脏功能辅助诊断和分析的基础，而左心室轮廓的提取则是正确分割心脏磁共振图像的关键。提出了一种提取心脏磁共振图像中左心室轮廓的方法。该方法首先采用一种自适应边缘保持平滑算法对心脏磁共振图像作平滑处理，接着采用K均值聚类算法对心脏磁共振图像作聚类分析，然后采用基于变分水平集方法的几何主动轮廓线模型提取左心室轮廓。实验表明，该方法能够克服心脏磁共振图像中的噪声和心脏周边组织的影响，而且具有较好的准确性和鲁棒性。

关键词: 医学图像分割, 心脏磁共振图像, 自适应平滑, K均值聚类, 主动轮廓模型, 水平集方法