一种基于ARG的肺结节良恶性度判定方法

摘要 基于CT影像的肺结节的良恶性识别是肺癌诊断的重要环节，针对这一问题，提出一种基于属性关系图(attributed relational graph， ARG)的肺结节良恶性度判定方法.该方法以肺结节CT图像块作为输入，利用ARG构建其特征结构，并从大量ARGs中挖掘与或图(and-or graph， AoG)作为肺结节类别识别模板，即肺结节良恶性度判定的依据.此外，为提高模板挖掘效率，该方法利用马尔可夫毯(Markov blanket， MB)发现算法去除图像中的冗余特征，降低ARG节点数量.实验结果表明，该方法对恶性肺结节的识别率达到90.12%，能够帮助正确、快速辨识与分析肺结节良恶性，具有一定的实用价值.
Abstract：Identification of benign and malignant pulmonary nodules is an important task during the diagnosis of lung cancer. Aimed for solving this problem， a method of measuring the malignancy of pulmonary nodules based on the attributed relational graph (ARG) was proposed. Feature structures of input lung nodule CT image patches were constructed with ARGs and the nodule category template was built by mining and-or graph (AoG) from ARGs in the proposed method. Moreover， Markov blanket discovering algorithm was applied for discriminative features selection to reduce the node number of ARGs， so the computational complexity of the graph matching for AoG mining was greatly reduced. Experimental results show that the recognition rate of malignant pulmonary nodules is up to 90.12%， thus the proposed method can help identify the benign and malignant pulmonary nodules accurately and rapidly.