论文部分内容阅读
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-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。
Objective To design and implement an automatic segmentation algorithm of articular cartilage for magnetic resonance imaging (MRI) images. Methods Two classes of support vector machines (SVMs) were used to classify the cartilage, tibiofibular cartilage and patellar cartilage using the global and local features of pixels respectively. Firstly, a threshold iteration method based on edge number feedback is proposed and the main edge of the image is extracted. Then, the extracted edges are identified according to the feature parameters and different bone-cartilage edges are marked. The training SVM classification The cartilage was preliminarily divided and the search space was reduced according to the anatomy of cartilage. Finally, morphological operation was used to optimize the initial segmentation results. Results The result of automatic segmentation showed that the contour of the cartilage was in good agreement with the original image. The average Dice’s coefficients of the cartilage, tibial cartilage and patellar cartilage were 0.80, 0.76 and 0.74, respectively, which were in good agreement with those of the manual segmentation. Conclusion The algorithm can accurately and quickly segment different cartilage tissue in MRI images.