【摘要】 乳腺癌是女性最常见的恶性肿瘤,而怎样利用超声弹性图像定量地诊断乳腺癌还是尚未解决的问题。该文根据乳腺肿块图像弹性信息,提取5个弹性特征来描述肿瘤的弹性属性;提取肿块的4个灰度共生矩阵特征来描述乳腺肿块的纹理特征。为进一步研究应用SVM分类器,对这些特征进行训练而完成分类,并且运用一致性、分类精度、ROC曲线及曲线下面积AUC对分类结果进行评估。利用超声弹性成像设备从实际病人采集数据,共采集了142例患者195个病灶的图像数据。实验结果表明,提取出的弹性特征分类性能良好,而支持向量机比较适合乳腺图像分类,分类精度比较高,有比较好的诊断价值。同时也发现,提取的灰度共生矩阵特征诊断价值比较低,参考意义较少。更多还原

【Abstract】 Breast cancers are the most common malignant tumors in women, and how to use ultrasound to diagnose breast cancers quantitatively is still an unsolved problem. This paper extracts five elastic features based on the elastography images of the breast tumors, furthers extract four features related to gray co-occurrence matrix to describe the texture of breast masses. we study the application of SVM classifier to classify these features, and uses the consistency, classification accuracy, ROC curve and AUC(area under the curve) to assess the classification results. we used ultrasound imaging technique to collect data from the actual patients, with the data of 195 lesions in 142 patients. Experimental results show that the classification performance of the elastic features is good, and the support vector machine is suitable for breast image classification, and its classification accuracy is high, which provides a good value for diagnosis. Meanwhile, it is found that the extracted features related to gray level co-occurrence matrix have a low diagnostic value.更多还原