[1]李语心1,赵艳娜2,谢荣理3,等.面向甲状腺结节良恶分类的cRes-GAN算法[J].机械与电子,2020,(04):6-10.
 ,,et al.A cRes-GAN Algorithm for Classification of Benign or Malignant Thyroid Nodules[J].Machinery & Electronics,2020,(04):6-10.
点击复制

面向 甲状腺结节良恶分类的cRes-GAN算法 ()
分享到:

机械与电子[ISSN:1001-2257/CN:52-1052/TH]

卷:
期数:
2020年04期
页码:
6-10
栏目:
设计与研究
出版日期:
2020-04-20

文章信息/Info

Title:
A cRes-GAN Algorithm for Classification of Benign  or Malignant Thyroid Nodules
文章编号:
1001- 2257(2020)04- 0006- 05
作者:
李语心赵艳娜谢荣理刘浚嘉付 庄王 尧张 俊费 健
1.上海交通大学机械系统与振动国家重点实验室,上海 200240;
2.上海市瑞金康复医院,上海 200023;
3.上海交通大学医学院附属瑞金医院,上海 200025
Author(s):
LIYuxinZHAOYannaXIERongliLIUJunjiaFUZhuangWANGYaoZHANGJunFEIJian
1.StateKeyLaboratoryofMechanicalSystemandVibration,ShanghaiJiaoTongUniversity,Shanghai200240,China;
2.ShanghaiRuijinRehabilitationHospital,Shanghai200023,China;
3.RuijinHospitalAffiliatedtoShanghaiJiaoTongUniversity,Shanghai200025,China
关键词:
甲状腺结节图像分类生成对抗网络cRes- GAN
Keywords:
thyroidnoduleimageclassificationGANcRes- GAN
分类号:
TP391.7
文献标志码:
A
摘要:
针对甲状腺结节良恶性分类问题,设计建立了条件限制残差生成对抗网络(cRes- GAN)算法。利用 DICOM 格式的1501份甲状腺结节数据建立了数据集,并且在该数据集上进行测试得到算法分类正确率为92.2%。将cRes- GAN 与 Hog+随机森林、ResNet18,Res18GAN,ACGAN 等其他4种算法相比,其分类正确率分别提升了24.8%,10.0%,12.6%和25.3%,分类效果得到了显著提升。所设计的算法可为医生的甲状腺结节良恶性诊断提供有效的辅助建议。
Abstract:

A new method of thyroid nodule malignancy classification algorithmwas proposed. Aiming at the classification problem of thyroid nodule DICOM files, a conditioned residual generative adversarial network (cRes-GAN) was designed. By using the discriminator of cRes-GAN to classify the ultrasound images of thyroid nodule, the accuracy arrives at 92.2%. Compared with the traditional image feature extraction method (Histogram of oriented gradient) followed by random forests classifier, residual networks, Res18GAN, and ACGAN, the accuracy of our method was significantly improved, of which the improvements are24.8%,10.0%,12.6%,25.3% repectively. cRes-GAN solved the problem of thyroid nodule classification problem, and can provide effective suggestions to doctors’ diagnosis.

参考文献/References:

[1] LIU TJ,XIESN,ZHANG Y K,etal.Featureselectionandthyroidnoduleclassificationusingtransfer learning[C]//2017IEEE14thInternationalSymposiumonBiomedicalImaging (ISBI2017).New York:IEEE,2017:1096- 1099.

[2] KRIZHEVSKY A,SUTSKEVERI,HINTON GE.Imagenetclassificationwithdeepconvolutionalneural networks[C]//Advancesin NeuralInformation ProcessingSystems,2012:1097- 1105.
[3] 叶晨,赵作鹏,马小平,等.基于 CNN 迁移学习的甲状腺结节检测方法[J].计算机工程与应用,2018,54(22):127- 132.
[4] GOODFELLOWI,POUGET ABADIEJ,MIRZA M,etal.Generativeadversarialnets[C]//AdvancesinNeural InformationProcessingSystems,2014:2672- 2680.
[5] DALALN,TRIGGSB.Histogramsoforientedgradientsforhumandetection[C]//2005IEEE Computer SocietyConferenceon Computer Visionand Pattern Recognition,2005:886 893.
[6] BREIMANL.Randomforests[J].MachineLearning,2001,45(1):5- 32.
[7] HEK M,ZHANGXY,RENSQ,etal.Deepresiduallearningforimagerecognition[C]//Proceedingsof theIEEEConferenceonComputerVisionandPattern Recognition,2016:770- 778.
[8] ODENA A,OLAH C,SHLENSJ.Conditionalimage synthesiswithauxiliaryclassifierGANs[C]//Proceedings
ofthe34thInternationalConferenceonMachineLearning,2017,70:2642- 2651.
[9] 竺烨.甲状腺结节医学图像处理及其在辅助诊断机器人中的应用[D].上海:上海交通大学,2019.

备注/Memo

备注/Memo:
收稿日期:2019- 12- 03
基金项目:国家自然基金面上项目(61973210);上海市科学技术委员会项目(17441901000);上海市黄浦区科研新项目(HKQ201810);医工交叉项目(YG2019ZDA17,ZH2018QNB23);上海市卫计委项目(201640230,20144Y0084)
作者简介:李语心 (1995-),女,上海人,硕士研究生,研究方向为机器学习;赵艳娜 (1985-),女,上海人,主治医师,研究方向为超声诊断及介入治疗;谢荣理 (1990-),男,上海人,住院医师,研究方向为机器人的临床应用;刘浚嘉 (1997-),男,辽宁大连人,硕士研究生,研究方向为强化学习;付 庄 (1972-),男,山东招远人,教授,研究方向为特种机器人与控制系统,通信作者;王 尧 (1988-),男,山西平遥人,博士、博士后,研究方向为医疗/穿刺手术机器人。
更新日期/Last Update: 2020-04-20