基于RoughSet的特征集融合PET/CT肺部肿瘤CAD模型

Lung Tumor CAD Model based on Rough Set with Feature-level Fusion in PET/CT Imaging

针对肺部肿瘤PET/CT感兴趣区域(ROI)在高维特征表示下存在着特征相关和维数灾难问题,提出了一种基于粗糙集特征集融合的PET/CT肺部肿瘤CAD模型。首先提取肺部肿瘤ROI的8维形状特征、7维灰度特征、3维Tamura纹理特征、56维GLCM特征和24维频域特征,得到98维特征矢量;然后基于遗传算法的知识约简方法和基于属性重要度的启发式算法对提取的特征集合分别进行特征级融合得到特征子集G1、G2、G3,A1、A2、A3,降低特征矢量的维数;再次利用网格寻优算法优化核函数的SVM作为分类器分别进行融合前和融合后的分类识别比较,基于遗传算法的特征集融合和基于属性重要度的特征集融合的分类识别比较2组实验;最后以2 000幅肺部肿瘤的PET/CT图像为原始数据,采用基于粗糙集特征集融合的肺部肿瘤PET/CT计算机辅助诊断模型对肺部肿瘤进行辅助诊断。实验结果表明,经过粗糙集特征集融合的肺部肿瘤诊断识别方法能有效提高肺部肿瘤的诊断正确率,一定程度上降低了特征之间的相关性。

Focusing on the issue that feature relevancy and dimension disaster problem in high-dimensional representation of PET/CT Lung tumor Region of Interesting（ROI）,a lung tumor CAD model was proposed based on support vector machine（SVM） with feature-level fusion in PET/CT.Firstly,98 dimension features were extracted from lung tumor ROI, including 8 dimensional shape features, 7 dimensional gray features, 3 dimensional tamura features, 56 dimensional GLCM features and 24 dimensional frequency features.Secondly, feature subsets G1, G2, G3 were obtained by using the knowledge reduction method based on genetic algorithm in feature-level fusion and feature subsets A1, A2, A3 were obtained by using heuristic algorithm based on attribute significance in feature-level fusion, reducing the dimension of feature vectors.Thirdly, using grid search algorithm to optimize the kernel function of the SVM as the classifier, compared classification before feature-level fusion and after feature-level fusion, compared classification between based on genetic algorithm in feature-level fusion and based on attribute significance in feature-level fusion in PET/CT.Finally, 2 000 PET/CT images of lung tumors as original data,and the lung tumor CAD model based on RoughSet with feature-level fusion in PET/CT was utilized to diagnose.The experimental results show that the method can effectively improve the accuracy of diagnosis of lung tumor, and increases the feature irrelevancy to a certain extent.