基于纹理特征的多序列MRI的肝硬化识别研究-论文.pdf

《基于纹理特征的多序列MRI的肝硬化识别研究-论文.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、中国医学影像技术2014年第3O卷第7期ChinJMedImagingTechnol，2014，Vol30，No7·11O5·●’．．影像技术学Multi—sequenceMRIclassificationofhepaticcirrhosisbasedontexturefeatureGUODong—mei，LIUHui。，SHAOYing，LINXiang—bo，LIUWen—hong。，JIHu(1．DepartmentofRadiology，theSecondAffiliatedHospitalofDalianMedicalUniversity，Dalian116027，Chin

2、a；2．FacultyofElectronicInformationandElectricalEngineering，Dalian116024，China；3．SchoolofElectronics&formationEngineering．ShanghaiDianJiUniversity。Shanghai201306。China)[Abstract]ObjectiveToinvestigatethediagnosticvalueofmulti—-sequencedynamicMRIforhepaticcirrhosisusingten—-foldcross—validationme

3、thodneuralnetworkclassifierbasedontexturefeature．MethodsT1WI，T2WI，arterialphase，portalvenousphaseandequilibriumphaseimagingweredividedintonormal，earlyandadvancedstagehepaticcirrhosisgroups．ROIoftheseimageswerecutmanually．Fifty-sixtexturefeatureswereextractedbygreylevelCO—occurrencematri—ces．Hep

4、atictissueswereclassifiedbyaBPclassifierbasedontenfoldcross—validationmethod．ResultsForclassificationofhepatictissueinall3groups，imagingofportalvenousphasewerethebest，andthetotalaccuracywas87．62(92／105)，T2W1weretheworst，withthetotalaccuracyof78．33(47／60)．T1WI，imagingofequilibriumphaseandarteria

5、lphasewereallbetterthanT2WI．ConclusionTenfoldcross_validationmethodneuralnetworkclassifiercanclassifynor—mal，earlyandadvancedstagehepaticcirrhosisonMRIbasedontexturefeature．Portalvenousphaseimagingmaybethefirstchoiceforclassificationofhepaticcirrhosisbasedonmulti—sequenceMRI．[Keywords]Livercirr

6、hosis；Texturefeature；Neuralnetwork；Magneticresonanceimaging；Multi—sequence基于纹理特征的多序列MRI的肝硬化识别研究郭冬梅，刘惠，邵莹。，林相波。，刘文红。，纪虎●(1．大连医科大学附属第二医院放射科，辽宁大连116027；2．大连理工大学电子信息与电气工程学部，辽宁大连116024；3．上海电机学院电子信息学院，上海201306)[摘要]目的采用基于纹理特征的十倍交叉验证法的神经网络分类器，探讨多序列MRI在肝硬化诊断识别中的价值。方法将5个序列MR图像(TIWI、T2WI、增强动脉期、门静脉期和平衡期)分成

7、正常肝脏组、较早期肝硬化组及中晚期肝硬化组，手动获取ROI；采用灰度共生矩阵提取ROI的56个纹理特征参数；采用十倍交叉验证法的BP网络分类器分别分类识别3组肝脏组织。结果门静脉期图像对正常肝脏、较早期肝硬化及中晚期肝硬化的分类效果最好，正确率为87．62(92／105)，T2w1分类效果最差，正确率为78．33(47／60)，T1wI、动脉期和平衡期图像居两者之间。结论采用基于纹理特征的十倍交叉验证法的神经网络分类器可以区分正常肝脏、较早期和中晚期肝硬化