基于CT图像统计纹理特征的甲状腺结节识别技术

发布时间：2018-02-24 06:44

本文关键词： 甲状腺结节灰度共生矩阵纹理特征支持向量机　出处：《航天医学与医学工程》2017年04期 　论文类型：期刊论文

【摘要】：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。
[Abstract]:Objective to explore the feasibility of differentiating benign and malignant thyroid nodules by statistical texture characteristics of gray level co-occurrence matrix and gray-scale gradient co-occurrence matrix on CT images. Methods 134 cases of thyroid nodules confirmed by operation and pathology were retrospectively collected. Manually extract the region of interest of the unilateral thyroid region with nodules. Calculate the statistical texture features of ROI and normalize them to [0]. Support Vector Machine (SVM) is used as a classifier. Results the accuracy of statistical texture features in differentiating benign and malignant thyroid nodules was 0.76, sensitivity 0.60, specificity 0.86, and area under the operating curve 0.81.Conclusion based on the results of this study, the accuracy of statistical texture features in the differential diagnosis of benign and malignant thyroid nodules is 0.86. Statistical texture features of grayscale co-occurrence matrix and grayscale gradient co-occurrence matrix, Thyroid CT images have a good classification effect for benign and malignant nodules.
【作者单位】：浙江大学生物医学工程教育部重点实验室;杭州医学院医学影像学院;Biodesign
【基金】：国家重点研发计划项目(2016YFC1306600) 浙江省卫计委科研项目(2015115991) 浙江省教育厅科研项目(Y201636958)
【分类号】：R581.3;R816.6;TP391.41

【参考文献】