中文摘要：

目的设计并实现一种针对核磁共振成像（MRI）图像的关节软骨自动分割算法。方法利用像素的整体与局部特征分别构建二分类支持向量机（support vector machine，SVM）分类器对股软骨、胫软骨及髌软骨进行自动分割。首先提出一种基于边缘数目反馈的Canny检测器阈值迭代法并利用该方法提取图像的主要边缘，随后根据特征参数对提取的边缘进行识别并标记出不同的骨一软骨边缘，利用训练的SVM分类器对软骨进行初步分割并根据软骨的解剖位置缩小搜索空间，最后利用形态学操作对初步分割结果进行优化。结果自动分割结果中软骨的形态轮廓与原始图像吻合效果好，股软骨、胫软骨及髌软骨的Dice＇s系数平均值分别为0．80、0．76、0．74，与手工分割结果具有较好的一致性。结论该算法能够准确、快速地分割出MRI图像中不同的软骨组织。

英文摘要：

Objective To explore the possibilities of automatic segmentation of articular cartilage in the MR images. Methods Three binary classifiers were built with edge-distance, intensity and other voxel features using support vector machine to segment the femur cartilage, tibia cartilage and patellar cartilage separately. Firstly, an iterative procedure based on the feedback of the number of main edges was designed to get an appropriate threshold for Canny operator and extract the main edges with it from MRI images. Secondly, the different edges with some feature parameters were identified, which distinguished different cartilages synchronously. To improve the speed of segmentation, the search space was reduced according to the cartilage edge and its anatomic location. Thirdly, the cartilage tissues were classifiers. Finally, morphological operations were used to make edge was smooth in automatic segmentation results and had a segmented preliminarily with different trained those results better. Results The cartilage good consistency with manual segmentation results. The mean Dice similarity coefficient of femoral cartilage, tibial cartilage and patellar cartilage was 0. 80, 0.76 and 0. 74, respectively. Conclusion The algorithm can segment the articular cartilage automatically from MR images more accurately and quickly.