论文部分内容阅读
目的探讨灰度共生矩阵和灰度梯度共生矩阵统计学纹理特征在CT图像上甲状腺结节良恶性鉴别的可行性。方法回顾性收集甲状腺结节经手术病理证实的CT图像134例,手动提取含结节的单侧甲状腺感兴趣区(region of interest,ROI)。计算ROI的统计学纹理特征并归一化到[0,1],支持向量机作为分类器,并结合留一交叉验证法来评价实验效果。结果统计学纹理特征在甲状腺结节良恶性鉴别中的准确率为0.76,敏感度0.60,特异性0.86和受试者操作曲线下面积为0.81。结论基于灰度共生矩阵和灰度梯度共生矩阵的统计法纹理特征,在甲状腺CT图像上对于结节的良恶性鉴别具有较好的分类效果。
Objective To explore the feasibility of differential diagnosis between benign and malignant thyroid nodules on the CT images by using the statistical texture features of gray level co-occurrence matrix and gray level gradient co-occurrence matrix. Methods Retrospectively collected 134 CT images of thyroid nodules confirmed by surgery and pathology, and manually extracted region of interest (ROI) with nodules. The statistical texture features of ROI were calculated and normalized to [0,1]. SVM was used as a classifier, and a cross validation method was used to evaluate the experimental results. Results The accuracy of statistical texture features in benign and malignant thyroid nodules was 0.76, with a sensitivity of 0.60, a specificity of 0.86, and an area under the operator’s operating curve of 0.81. Conclusion Based on the statistical texture features of the gray level co-occurrence matrix and the grayscale gradient co-occurrence matrix, it has a good classification effect on benign and malignant nodules in thyroid CT images.