结合候选区域距离度量学习与CNN分类回归联合的左心室检测
|
摘要
心脏MRI左心室自动检测在心脏疾病计算机辅助诊断中具有重要价值,针对左心室候选区域与周边组织分布复杂而导致区分度低的问题,提出结合候选区域2级距离度量学习与CNN分类回归联合学习的左心室检测方法.在候选区域生成阶段,利用超像素产生初始区域并合并为中间区域,设计有监督的2级距离度量学习算法,融合中间区域来构建目标候选区域;在检测阶段,以CNN分类与回归联合学习的方式定位候选区域,并设计难例样本挖掘策略对模型进行微调,以缓解样本不均衡问题.将该方法与扩展的4种变体方法(改变或舍弃部分模块)在公开心脏图谱数据集(CAP)上进行了实验,结果表明该方法中各模块设置具有合理性;与FastR-CNN和基于SSAE方法的检测结果相比,该方法取得了较高的检测精度.
Automatic detection of left ventricle(LV) in cardiac MRI is of great value for computer-aided diagnosis of heart disease. To solve the low discrimination problem between left ventricular candidate regions and surrounding tissue, a left ventricular detection method is proposed by combining candidate region two-level distance metric learning and CNN classification and regression joint learning. In the candidate region generation stage, the super-pixel method is employed to generate the initial region and further merged into the intermediate region. The supervised two-level distance metric learning algorithm is designed to fuse the intermediate regions to construct the target candidate regions. In the detection stage, the approach of joint learning with CNN classification and regression is employed to locate candidate regions, and a hard negative mining strategy is added for tuning the model adaptive to the sample imbalance problem. The proposed method and the extended four variant methods(changing or discarding some modules) are performed on the Cardiac Atlas Project(CAP) data set and the results validate the reasonability of module settings in this method. Further experiments show that the proposed method achieves higher detection accuracy in comparison with the Fast R-CNN and SSAE-based methods.
Automatic detection of left ventricle(LV) in cardiac MRI is of great value for computer-aided diagnosis of heart disease. To solve the low discrimination problem between left ventricular candidate regions and surrounding tissue, a left ventricular detection method is proposed by combining candidate region two-level distance metric learning and CNN classification and regression joint learning. In the candidate region generation stage, the super-pixel method is employed to generate the initial region and further merged into the intermediate region. The supervised two-level distance metric learning algorithm is designed to fuse the intermediate regions to construct the target candidate regions. In the detection stage, the approach of joint learning with CNN classification and regression is employed to locate candidate regions, and a hard negative mining strategy is added for tuning the model adaptive to the sample imbalance problem. The proposed method and the extended four variant methods(changing or discarding some modules) are performed on the Cardiac Atlas Project(CAP) data set and the results validate the reasonability of module settings in this method. Further experiments show that the proposed method achieves higher detection accuracy in comparison with the Fast R-CNN and SSAE-based methods.
引文
[1]Zhong L, Zhang J M, Zhao X D, et al. Automatic localization of the left ventricle from cardiac cine magnetic resonance imaging:anewspectrum-basedcomputer-aidedtool[J].PLoS ONE, 2014, 9(4):e92382
[2]Mousa D, Zayed N, Yassine I A. Automatic cardiac MRI localizationmethod[C]//ProceedingsoftheCairoInternational BiomedicalEngineeringConference.LosAlamitos:IEEE Computer Society Press, 2014:153-157
[3]Tan L K, Liew Y M, Lim E, et al. Automatic localization of the leftventricularbloodpoolcentroidinshortaxiscardiaccine MR images[J]. Medical&Biological Engineering&Computing, 2018, 56(6):1053-1062
[4]Kurkure U, Pednekar A, Muthupillai R, et al. Localization and segmentationofleftventricleincardiaccine-MRimages[J].IEEETransactionsonBiomedicalEngineering,2009,56(5):1360-1370
[5]MarinoM,VeronesiF,TarroniG,etal.Fullyautomatedassessment of left ventricular volumes, function and mass from cardiac MRI[C]//Proceedings of the 41st Computing in Cardiology Conference. Los Alamitos:IEEE Computer Society Press,2014:109-112
[6]HelwanA,OzsahinDU.Slidingwindowbasedmachine learning system for the left ventricle localization in MR cardiac images[J]. Applied Computational Intelligence and Soft Computing, 2017, 2017:Article No.3048181
[7]Wang Xuchu, Niu Yanmin, Zhao Guangjun, et al. Combining region proposals and deep SSAE learnt features for detecting leftventricleincardiacMRimages[J].JournalofComputer-AidedDesign&ComputerGraphics,2018,30(3):424-435(in Chinese)(王旭初,牛彦敏,赵广军,等.融合候选区域提取与SSAE深度特征学习的心脏MR图像左心室检测[J].计算机辅助设计与图形学学报, 2018, 30(3):424-435)
[8]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies foraccurateobjectdetectionandsemanticsegmentation[C]//Proceedings of the IEEE Conference on Computer Vision and PatternRecognition.LosAlamitos:IEEEComputerSociety Press, 2014:580-587
[9]Girshick R. Fast R-CNN[C]//Proceedings of the IEEE InternationalConferenceonComputerVision.LosAlamitos:IEEE Computer Society Press, 2015:1440-1448
[10]Ren S Q, He K M, Girshick R, et al. Faster R-CNN:towards real-timeobjectdetectionwithregionproposalnetworks[J].IEEETransactionsonPatternAnalysisandMachineIntelligence, 2017, 39(6):1137-1149
[11]Dai J F, Li Y, He K M, et al. R-FCN:object detection via region-basedfullyconvolutionalnetworks[C]//Proceedingsof theIEEEComputerSocietyConferenceonComputerVision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:379-387
[12]LinT Y, DollárP, Girshick R B, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE ConferenceonComputerVisionandPatternRecognition.Los Alamitos:IEEE Computer Society Press, 2017:936-944
[13]Redmon J, Divvala S, Girshick R, et al. You only look once:unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:779-788
[14]RedmonJ,FarhadiA.YOLO9000:better,faster,stronger[C]//Proceedings of the IEEE Conference on Computer Vision and PatternRecognition.LosAlamitos:IEEEComputerSociety Press, 2017:6517-6525
[15]Liu W, Anguelov D, Erhan D, et al. SSD:single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision. Heidelberg:Springer, 2016:21-37
[16]Uijlings J R R, van de Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2):154-171
[17]Lee T, Fidler S, Dickinson S. Learning to combine mid-level cuesforobjectproposalgeneration[C]//Proceedingsofthe IEEEInternationalConferenceonComputerVision.Los Alamitos:IEEE Computer Society Press, 2016:1680-1688
[18]Rantalankila P, Kannala J, Rahtu E. Generating object segmentation proposals using global and local search[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2014:2417-2424
[19]Sun N, Jiang F, Yan H C, et al. Proposal generation method for object detection in infrared image[J]. Infrared Physics&Technology, 2017, 81:117-127
[20]Arbeláez P, Pont-Tuset J, Barron J T, et al. Multiscale combinatorial grouping[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2014:328-335
[21]Carreira J, Sminchisescu C. CPMC:automatic object segmentation using constrained parametric min-cuts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7):1312-1328
[22]Humayun A, Li F X, Rehg J M. RIGOR:reusing inference in graph cuts for generating object regions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press, 2014:336-343
[23]Xiao Y, Lu C W, Tsougenis E, et al. Complexity-adaptive distancemetricforobjectproposalsgeneration[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2015:778-786
[24]Achanta R, Shaji A, Smith K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on PatternAnalysisandMachineIntelligence,2012,34(11):2274-2282
[25]Wang F, Sun J M. Survey on distance metric learning and dimensionalityreductionindatamining[J].DataMiningand Knowledge Discovery, 2015, 29(2):534-564
[26]Davis J V, Kulis B, Jain P, et al. Information-theoretic metric learning[C]//Proceedings of the 24th International Conference on Machine Learning. Madison:Omnipress, 2007:209-216
[27]QiGJ,TangJH,ZhaZJ,etal.Anefficientsparsemetric learninginhigh-dimensionalspaceviaL1-penalized log-determinantregularization[C]//Proceedingsofthe26th AnnualInternationalConferenceonMachineLearning.New York:ACM Press, 2009:841-848
[28]Bell S, Lawrence Zitnick C, Bala K, et al. Inside-outside net:detectingobjectsincontextwithskippoolingandrecurrent neural networks[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press, 2016:2874-2883
[29]Fonseca C G, Backhaus M, Bluemke D A, et al. The cardiac atlas project-an imaging database for computational modeling andstatisticalatlasesoftheheart[J].Bioinformatics,2011,27(16):2288-2295
[2]Mousa D, Zayed N, Yassine I A. Automatic cardiac MRI localizationmethod[C]//ProceedingsoftheCairoInternational BiomedicalEngineeringConference.LosAlamitos:IEEE Computer Society Press, 2014:153-157
[3]Tan L K, Liew Y M, Lim E, et al. Automatic localization of the leftventricularbloodpoolcentroidinshortaxiscardiaccine MR images[J]. Medical&Biological Engineering&Computing, 2018, 56(6):1053-1062
[4]Kurkure U, Pednekar A, Muthupillai R, et al. Localization and segmentationofleftventricleincardiaccine-MRimages[J].IEEETransactionsonBiomedicalEngineering,2009,56(5):1360-1370
[5]MarinoM,VeronesiF,TarroniG,etal.Fullyautomatedassessment of left ventricular volumes, function and mass from cardiac MRI[C]//Proceedings of the 41st Computing in Cardiology Conference. Los Alamitos:IEEE Computer Society Press,2014:109-112
[6]HelwanA,OzsahinDU.Slidingwindowbasedmachine learning system for the left ventricle localization in MR cardiac images[J]. Applied Computational Intelligence and Soft Computing, 2017, 2017:Article No.3048181
[7]Wang Xuchu, Niu Yanmin, Zhao Guangjun, et al. Combining region proposals and deep SSAE learnt features for detecting leftventricleincardiacMRimages[J].JournalofComputer-AidedDesign&ComputerGraphics,2018,30(3):424-435(in Chinese)(王旭初,牛彦敏,赵广军,等.融合候选区域提取与SSAE深度特征学习的心脏MR图像左心室检测[J].计算机辅助设计与图形学学报, 2018, 30(3):424-435)
[8]Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies foraccurateobjectdetectionandsemanticsegmentation[C]//Proceedings of the IEEE Conference on Computer Vision and PatternRecognition.LosAlamitos:IEEEComputerSociety Press, 2014:580-587
[9]Girshick R. Fast R-CNN[C]//Proceedings of the IEEE InternationalConferenceonComputerVision.LosAlamitos:IEEE Computer Society Press, 2015:1440-1448
[10]Ren S Q, He K M, Girshick R, et al. Faster R-CNN:towards real-timeobjectdetectionwithregionproposalnetworks[J].IEEETransactionsonPatternAnalysisandMachineIntelligence, 2017, 39(6):1137-1149
[11]Dai J F, Li Y, He K M, et al. R-FCN:object detection via region-basedfullyconvolutionalnetworks[C]//Proceedingsof theIEEEComputerSocietyConferenceonComputerVision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:379-387
[12]LinT Y, DollárP, Girshick R B, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE ConferenceonComputerVisionandPatternRecognition.Los Alamitos:IEEE Computer Society Press, 2017:936-944
[13]Redmon J, Divvala S, Girshick R, et al. You only look once:unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2016:779-788
[14]RedmonJ,FarhadiA.YOLO9000:better,faster,stronger[C]//Proceedings of the IEEE Conference on Computer Vision and PatternRecognition.LosAlamitos:IEEEComputerSociety Press, 2017:6517-6525
[15]Liu W, Anguelov D, Erhan D, et al. SSD:single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision. Heidelberg:Springer, 2016:21-37
[16]Uijlings J R R, van de Sande K E A, Gevers T, et al. Selective search for object recognition[J]. International Journal of Computer Vision, 2013, 104(2):154-171
[17]Lee T, Fidler S, Dickinson S. Learning to combine mid-level cuesforobjectproposalgeneration[C]//Proceedingsofthe IEEEInternationalConferenceonComputerVision.Los Alamitos:IEEE Computer Society Press, 2016:1680-1688
[18]Rantalankila P, Kannala J, Rahtu E. Generating object segmentation proposals using global and local search[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2014:2417-2424
[19]Sun N, Jiang F, Yan H C, et al. Proposal generation method for object detection in infrared image[J]. Infrared Physics&Technology, 2017, 81:117-127
[20]Arbeláez P, Pont-Tuset J, Barron J T, et al. Multiscale combinatorial grouping[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2014:328-335
[21]Carreira J, Sminchisescu C. CPMC:automatic object segmentation using constrained parametric min-cuts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7):1312-1328
[22]Humayun A, Li F X, Rehg J M. RIGOR:reusing inference in graph cuts for generating object regions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press, 2014:336-343
[23]Xiao Y, Lu C W, Tsougenis E, et al. Complexity-adaptive distancemetricforobjectproposalsgeneration[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Los Alamitos:IEEE Computer Society Press, 2015:778-786
[24]Achanta R, Shaji A, Smith K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on PatternAnalysisandMachineIntelligence,2012,34(11):2274-2282
[25]Wang F, Sun J M. Survey on distance metric learning and dimensionalityreductionindatamining[J].DataMiningand Knowledge Discovery, 2015, 29(2):534-564
[26]Davis J V, Kulis B, Jain P, et al. Information-theoretic metric learning[C]//Proceedings of the 24th International Conference on Machine Learning. Madison:Omnipress, 2007:209-216
[27]QiGJ,TangJH,ZhaZJ,etal.Anefficientsparsemetric learninginhigh-dimensionalspaceviaL1-penalized log-determinantregularization[C]//Proceedingsofthe26th AnnualInternationalConferenceonMachineLearning.New York:ACM Press, 2009:841-848
[28]Bell S, Lawrence Zitnick C, Bala K, et al. Inside-outside net:detectingobjectsincontextwithskippoolingandrecurrent neural networks[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Los Alamitos:IEEE Computer Society Press, 2016:2874-2883
[29]Fonseca C G, Backhaus M, Bluemke D A, et al. The cardiac atlas project-an imaging database for computational modeling andstatisticalatlasesoftheheart[J].Bioinformatics,2011,27(16):2288-2295