运用超声造影特征选择的淋巴结良恶性鉴别

Classification of Benign and Malignant Lymph Nodes by Using Feature Selection of Contrast-Enhanced Ultrasound

提取淋巴结超声造影(CEUS)图像的影像组学量化特征可用于淋巴结良恶性的计算机辅助诊断。由于大量特征之间存在冗余和干扰信息,需借助特征选择技术进行特征降维,以获得更具鉴别能力的特征子集。利用实时压缩感知算法进行CEUS视频中淋巴结病灶的运动补偿,提取时域与空域特征。运用最小绝对压缩(LASSO)法、支持向量机回归特征法(SVM-RFE)、Fisher准则法三种特征选择方法,对特征进行降维。运用支持向量机进行交叉验证,得到分类结果。相对原始特征,三种特征选择方法得到的特征子集的分类性能均有提升。其中,运用LASSO进行降维的效果最好,分类的准确率、精度、敏感性、特异性和约登指数分别达到98.5%、100%、97.1%、100%和97.1%,相较全体特征的分类结果分别提升11.4%、14.8%、15.0%、14.3%和29.2%。结果表明,对影像组学量化特征的降维能够筛选出更具鉴别能力的特征子集,从而提升计算机辅助诊断的性能。

Extracting quantitative features from contrast-enhanced ultrasound（ CEUS） images by using radiomics is useful for computer aided diagnosis of benign and malignant lymph nodes. Because of the redundancy and interference information among a large number of features,feature selection techniques are needed to perform feature dimensionality reduction to obtain feature subsets with more discriminative capabilities. The motion compensation of lymph node lesions in a CEUS video is performed by using real-time compressive sensing algorithm,and then the time domain and space domain features are extracted. Subsequently,dimensionality reduction of features is conducted by using three feature selection methods,namely the least absolute shrinkage and selection operator（ LASSO）,recursive feature elimination based on support vector machine（ SVM-RFE）,and Fisher criteria. The support vector machine is employed to do cross validation and get classification result. Relative to the original features,the classification performance of the feature subsets obtained by the three feature selection methods has been improved,and the effect by using LASSO is the best. The accuracy,precision,sensitivity,specificity and Youden index are 98. 5%,100%,97. 1%,100%and 97. 1%,respectively,which are increased by 11. 4%,14. 8%,15. 0%,14. 3% and 29. 2% compared to the results of features in the whole set. These results demonstrate that the feature selection methods can achieve subset containing the most discriminative features extracted by using radiomics,which can improve the performance of computer aided diagnosis.