资源描述:
《基于密集特征匹配的胸片肺野自动分割》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、JSouthMedUniv,2016,36(1):61-66doi10.3969/j.issn.1673-4254.2016.01.11·61·技术方法ࠎႿૡࠢหᆘྣோ۩ٳሱ့٫佘广南,陈莹胤,钟丽明,阳维,冯前进南方医科大学生物医学工程学院,广东广州510515摘要:目的X线胸片中肺野的准确分割是胸片图像自动分析的必要步骤。本文采用局部特征的密集匹配和标号融合进行胸片肺野的自动分割。方法对于输入的待分割胸片,基于每个像素点提取密集SIFT描述子和图像块作为局部特征,采用密集匹配直接在整个参考图像特征集合中快速搜索
2、近邻;密集匹配分为受限的随机初始化、近邻场传播和受限的随机搜索三步,并数次迭代后两步。利用匹配得到的近邻场,提取标号图像块并进行加权,权重为匹配的相似度,最后重组为肺野的概率图,经阈值化处理即可得到肺野的分割。结果在公开的JSRT胸片图像数据集上进行测试,本文方法的Jaccard指标可达95.5%。结论利用局部特征的密集匹配和标号融合能取得准确性高且稳定的胸片肺野分割效果,并且优于当前最好的胸片肺野分割方法。关键词:局部特征;密集匹配;标号融合;肺野分割;胸片Automaticsegmentationoflungfiel
3、dsinchestradiographsbasedondensematchingoflocalfeaturesSHEGuangnan,CHENYingyin,ZHONGLiming,YANGWei,FENGQianjinSchoolofBiomedicalEngineering,SouthernMedicalUniversity,Guangzhou510515,ChinaAbstract:ObjectiveAccuratesegmentationoflungfieldsinchestradiographs(CXR)isv
4、eryusefulforautomaticanalysisofCXR.Inthiswork,weproposetousedensematchingoflocalfeaturesandlabelfusiontoautomaticallysegmentthelungfieldsinCXR.MethodsForaninputCXR,thedenseScaleInvariantFeatureTransform(SIFT)descriptorsandrawimagepatcheswereextractedasthelocalfea
5、turesforeachpixel.Thenearestneighborsofthelocalfeatureswerethenquicklysearchedbydensematchingdirectlyfromthewholefeaturedatasetofthereferenceimages.Thedensematchingincludedthreesteps:limitedrandominitialization,propagationofnearestneighborfield,andlimitedrandomse
6、arch,withiterationofthelasttwostepsforseveraltimes.Thelabelimagepatchesforeachpixelwereextractedaccordingtothenearestneighborfieldandweightedbythematchingsimilarity.Finally,theweightedlabelpatcheswererearrangedasthelabelclassprobabilityimageoftheinputCXR,fromwhic
7、hthresholdswereobtainedforsegmentationofthelungfields.ResultsTheJaccardindexoftheproposedmethodreached95.5%onthepublicJSRTdataset.ConclusionAhighaccuracyandrobustnesscanbeobtainedbyadoptingdensematchingoflocalfeaturesandlabelfusiontosegmentthelungfieldsinCXR,andt
8、heresultisbetterthanthatofcurrentsegmentationmethod.Keywords:localfeature;densematching;labelfusion;lungsegmentation;chestradiograph[1]X线胸片是目前临床上广泛使用的医学成像技术,目前