基于语义属性的肺结节良恶性分类

现有肺结节良恶性计算机辅助诊断的依据通常为肺部CT图像的底层特征,而临床医生的诊断依据为高级语义特征.为克服这种图像底层特征和高级语义特征之间的不一致性,提出一种基于语义属性的肺结节良恶性判别方法.首先,利用阈值概率图方法提取肺结节图像;其次,一方面提取肺结节图像的形状、灰度、纹理、大小和位置等底层特征,组成样本特征集.另一方面,根据专家对肺结节属性的标注,提取结节属性集;然后,根据特征集和属性集建立属性预测模型,实现两者之间的映射;最后,利用预测的属性进行肺结节的良恶性分类.LIDC数据库上的实验结果表明所提方法具有较高的分类精度和AUC值.

Benign or Malignant Classification of Lung Nodules Based on Semantic Attributes

The current computer aided diagnosis system classifies benign or malignant lung nodules mainly according to the low-level features of lung CT images.However,clinicians use the high-level semantic features of lung CT images.To overcome the inconsistency between the low-level features and high-level semantic features,a new approach of benign or malignant lung nodules classification based on semantic attributes is proposed.Firstly,lung nodule images are extracted using the threshold probability-map method.Secondly,on the one hand,some features including shape,gray,texture,size and position are extracted from lung nodule images to constitute the low-level feature set;on the other hand,according to the experts' annotation of lung nodules,the attributes are extracted to constitute the high-level attribute set.Thirdly,attribute prediction models are built to map the low-level features to the high-level attributes.Finally,the benign or malignant classification of lung nodules is performed using the predicted attributes.Experimental results on the LIDC dataset show that the proposed classification method possesses high classification accuracy and AUC value.