目的 比较动态对比度增强磁共振成像(dynamic contrast-enhanced magnetic resonance imaging,DCE-MRI)图像的形态、纹理和时间强度曲线(time intensity curve,TIC)特征对乳腺病灶良恶性的诊断效果,讨论DCE-MRI图像特征的计算机辅助诊断价值.方法 测量224个乳腺病灶样本(良性样本82个,恶性样本142个)的12个形态学特征、56个基于灰度共生矩阵(gray level co-occurrence matrix,GLCM)的纹理特征以及11个TIC特征,采用平均平方距离准则和SVM分类器评估这三类特征的良恶性分辨能力.结果 反映病灶血流动力学特性的TIC特征的分类性能最优(SE 0.9366,SP 0.8293,AUC 0.9495);纹理特征次之(SE 0.9225,SP 0.7195,AUC 0.8835);形态学特征效果最差(SE 0.8451,SP 0.6951,AUC 0.8079).研究发现,在上述基础上融合三类特征可优化分类性能.最终结合平滑度、紧致度、熵等9个特征参数进行诊断,对乳腺病灶良恶性的分辨效果最好,AUC达0.9642.结论 DCE-MRI的TIC特征对恶性乳腺病灶具有较高的灵敏度,可以提高乳腺计算机辅助诊断的恶性病灶检出率.综合分析形态、纹理和TIC特征可以提高病灶的诊断特异度,降低良性病灶的误诊率.%To discuss the applicability of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) features in diagnosis of breast lesions between benign and malignant by comparing the classification performances of morphology, texture and time intensity curves ( TIC ). Methods Twelve morphological features, 56 gray level co-occurrence matrix (GLCM) based texture features and 11 features of TIC are calculated for the data of 224 breast lesions ( benign 82, malignant 142). To evaluate the performance of discriminating malignant from benign lesions, these features are studied based on average square distance criteria and SVM classifier. Results The TIC features ( sensitivity 0. 9366 , specificity 0. 8293 , AUC 0. 9495 )rnperform best and the performance of texture features (sensitivity 0. 9225, specificity 0. 7195, AUC 0. 8835) are better than morphological features ( sensitivity 0. 8451, specificity 0. 6951, AUC 0. 8079 ). The combination of the 9 features ( smoothness , compactness , entropy, etc. ) performs optimally ( AUC 0.9642 ). Conclusions The TIC features are of high sensitivity formalignancy and improve the identification rate of malignant lesions in breast computer-aided diagnosis. The comprehensive analysis of the morphological features, texture features and TIC features can reduce the misdiagnosis rate and improve the specificity of malignant lesions at the same time.
展开▼
