甲状腺良恶性结节超声二维灰阶图像鉴别的数学模型比较研究  

Comparative study on mathematical models of ultrasonic two-dimensional grayscale image identification of benign and malignant thyroid nodules

作　　者：隋鑫[1] 解朋[2] 刘宗杰 赵颖[1] 陈思美 赵冰歆 刘佳 张屹[3] SUI Xin;XIE Peng;LIU Zongjie;ZHAO Ying;CHEN Simei;ZHAO Bingxin;LIU Jia;ZHANG Yi(Department of Ultrasound Medicine,the Third Hospital of Hebei Medical University,Hebei Province,Shijiazhuang 050051,China;Department of Nuclear Medicine,the Third Hospital of Hebei Medical University,Hebei Province,Shijiazhuang 050051,China;School of Sinences,Heibei University of Science and Technology,Hebei Province,Shijiazhuang 050018,China)

机构地区：[1]河北医科大学第三医院超声医学科,河北石家庄050051 [2]河北医科大学第三医院核医学科,河北石家庄050051 [3]河北科技大学理学院,河北石家庄050018

出　　处：《中国医药导报》2022年第16期156-160,共5页China Medical Herald

基　　金：河北省重点研发计划项目(18277719D);河北省自然科学基金资助项目(A2019208336)。

摘　　要：目的构建并比较甲状腺良恶性结节超声二维灰阶图像的多种数学模型。方法收集河北医科大学第三医院2015年6月至2021年6月行超声检查发现甲状腺结节的504例患者。术中共切取546个甲状腺结节样本行病理学检查,其中良性结节415个,恶性结节131个。根据甲状腺结节超声二维图像的灰度值,找到区分良恶性的显著征象;通过数学模型建立多种分类器,比较不同分类器显示良恶性结节的精度和覆盖度,找到最优的数学分类模型。结果通过用良恶性的显著特征构建Fisher分类、回归分类、贝叶斯分类和Libsvm分类。Libsvm分类诊断甲状腺结节良恶性的精度和覆盖度最优,诊断甲状腺良性结节的精确度为93.8%,覆盖度为78.9%;甲状腺恶性结节精确度为81.8%,覆盖度为94.7%。结论Libsvm分类器可以快捷精准地判断甲状腺结节良恶性,进一步促进甲状腺结节的临床诊断及治疗。Objective To establish and compare various mathematical models of ultrasonic two-dimensional grayscale image of benign and malignant thyroid nodules.Methods A total of 504 patients diagnosed with thyroid nodule in the Third Hospital of Hebei Medical University from June 2015 to June 2021 were collected.A total of 546 thyroid nodule samples were cut for pathological examination,including 415 benign nodules and 131 malignant nodules.According to the gray value of two-dimensional ultrasonic image of thyroid nodule,the significant characteristics of differentiating benign and malignant thyroid nodules were found;a variety of classifiers were established by mathematical model,and the best mathematical classification model was found by comparing the accuracy and coverage of different classifiers.Results Using significant features of benign and malignant nodules build Fisher classification,regression classification,Bayesian classification,and Libsvm classification.Libsvm classification of benign and malignant thyroid nodules has the best accuracy and coverage,and the accuracy of benign thyroid nodules was 93.8%,coverage was 78.9%;and the accuracy of malignant thyroid nodules was 81.8%,coverage was 94.7%.Conclusion Libsvm classifier can quickly and accurately determine the benign and malignant thyroid nodules,and further promote the clinical diagnosis and treatment of thyroid nodules.

关 键 词：甲状腺结节 超声 Libsvm分类 数学模型 

分 类 号：R581[医药卫生—内分泌]