《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (1): 316-324.DOI: 10.11772/j.issn.1001-9081.2021010200
收稿日期:2021-02-03
									
				
											修回日期:2021-04-28
									
				
											接受日期:2020-04-29
									
				
											发布日期:2021-05-07
									
				
											出版日期:2022-01-10
									
				
			通讯作者:
					杨光
							作者简介:吴洁(1995—),女,江苏扬州人,硕士研究生,主要研究方向:机器学习、医学图像处理基金资助:
        
                                                                                                                            Jie WU1,2, Shitian ZHANG3, Haibin XIE1,2, Guang YANG1,2( )
)
			  
			
			
			
                
        
    
Received:2021-02-03
									
				
											Revised:2021-04-28
									
				
											Accepted:2020-04-29
									
				
											Online:2021-05-07
									
				
											Published:2022-01-10
									
			Contact:
					Guang YANG   
							About author:WU Jie, born in 1995, M. S. candidate. Her research interests include machine learning, medical image processing.Supported by:摘要:
针对大量数据手工标记的繁重性和单一影像中心磁共振成像(MRI)数据的有限性问题,提出了一种利用多影像中心有标签与无标签MRI数据的用于磁共振的半监督学习(MRSSL)方法,并将其应用在膝盖异常分类任务中。首先,运用了数据扩增方法来提供模型所需的归纳偏置;接着,融合了分类损失项和一致性损失项来约束人工神经网络并使之从数据中提取出具有辨别力的特征;然后,将这些特征用于MRI膝盖异常分类。此外,也提出了对应的仅利用有标签数据的完全监督学习(MRSL)方法。在给出同样的有标签样本时,将MRSL与MRSSL进行了比较,结果表明MRSSL的模型分类性能与泛化性能明显优于MRSL。最后,将MRSSL与其他半监督学习方法进行了比较。结果表明数据扩增在性能提升中起到了重要作用,并且MRSSL凭借更强的MRI数据包容性取得了最优的膝盖异常分类性能。
中图分类号:
吴洁, 张师天, 谢海滨, 杨光. 基于多影像中心磁共振成像数据的半监督膝盖异常分类[J]. 计算机应用, 2022, 42(1): 316-324.
Jie WU, Shitian ZHANG, Haibin XIE, Guang YANG. Semi-supervised knee abnormality classification based on multi-imaging center MRI data[J]. Journal of Computer Applications, 2022, 42(1): 316-324.
| 数据集 | 膝盖正常 | 膝盖异常 | 合计 | |||
|---|---|---|---|---|---|---|
| 前交叉韧带撕裂 | 半月板撕裂 | 前交叉韧带与半月板都撕裂 | 其他膝盖异常 | |||
| MRNet训练集 | 217 | 83 | 272 | 125 | 433 | 1 130 | 
| MRNet验证集 | 25 | 23 | 21 | 31 | 20 | 120 | 
| 本文训练集 | 193 | 59 | 248 | 101 | 409 | 1 010 | 
| 本文验证集 | 24 | 24 | 24 | 24 | 24 | 120 | 
表1 MRNet数据集的原始训练集、验证集与本文实验中训练集、验证集的膝盖情况分布
Tab. 1 Distribution of original training set and validation set of MRNet dataset and training set and validation set used in the paper
| 数据集 | 膝盖正常 | 膝盖异常 | 合计 | |||
|---|---|---|---|---|---|---|
| 前交叉韧带撕裂 | 半月板撕裂 | 前交叉韧带与半月板都撕裂 | 其他膝盖异常 | |||
| MRNet训练集 | 217 | 83 | 272 | 125 | 433 | 1 130 | 
| MRNet验证集 | 25 | 23 | 21 | 31 | 20 | 120 | 
| 本文训练集 | 193 | 59 | 248 | 101 | 409 | 1 010 | 
| 本文验证集 | 24 | 24 | 24 | 24 | 24 | 120 | 
| 等级 | 颜色变换 | 仿射变换 | ||||
|---|---|---|---|---|---|---|
| 亮度 | 对比度 | 水平翻转 | 缩放 | 旋转 | 平移 | |
| 0 | 0.0 | 0.0 | 0.5 | 0.0 | 0 | 0.10 | 
| 1 | 0.1 | 0.1 | 0.5 | 0.3 | 15 | 0.05 | 
| 2 | 0.2 | 0.2 | 0.5 | 0.3 | 25 | 0.10 | 
| 3 | 0.3 | 0.3 | 0.5 | 0.3 | 35 | 0.15 | 
| 4 | 0.4 | 0.4 | 0.5 | 0.3 | 45 | 0.20 | 
表2 MRAugment的5种扩增等级
Tab. 2 Five augmentation levels of MRAugment
| 等级 | 颜色变换 | 仿射变换 | ||||
|---|---|---|---|---|---|---|
| 亮度 | 对比度 | 水平翻转 | 缩放 | 旋转 | 平移 | |
| 0 | 0.0 | 0.0 | 0.5 | 0.0 | 0 | 0.10 | 
| 1 | 0.1 | 0.1 | 0.5 | 0.3 | 15 | 0.05 | 
| 2 | 0.2 | 0.2 | 0.5 | 0.3 | 25 | 0.10 | 
| 3 | 0.3 | 0.3 | 0.5 | 0.3 | 35 | 0.15 | 
| 4 | 0.4 | 0.4 | 0.5 | 0.3 | 45 | 0.20 | 
| 方法 | MRAugment等级/等级组合 | 验证集AUC | 测试集AUC | ||||
|---|---|---|---|---|---|---|---|
| 平均值 | 标准差 | 中位数 | 平均值 | 标准差 | 中位数 | ||
| MRSL | 0.936 | 0.008 | 0.938 | 0.856 | 0.012 | 0.858 | |
| 0.956 | 0.010 | 0.960 | 0.916 | 0.012 | 0.914 | ||
| 0.965 | 0.009 | 0.965 | 0.919 | 0.013 | 0.915 | ||
| 0.961 | 0.013 | 0.964 | 0.906 | 0.013 | 0.909 | ||
| MRSSL | 0.947 | 0.010 | 0.946 | 0.928 | 0.012 | 0.926 | |
| 0.954 | 0.009 | 0.957 | 0.939 | 0.006 | 0.939 | ||
| 0.960 | 0.008 | 0.960 | 0.946 | 0.010 | 0.949 | ||
| 0.951 | 0.005 | 0.952 | 0.944 | 0.015 | 0.944 | ||
| 0.951 | 0.011 | 0.950 | 0.942 | 0.009 | 0.945 | ||
| 0.951 | 0.011 | 0.952 | 0.942 | 0.008 | 0.941 | ||
| 0.949 | 0.007 | 0.947 | 0.933 | 0.010 | 0.931 | ||
表3 MRSL与MRSSL分别在不同MRAugment扩增等级与等级组合下的AUC的统计情况
Tab. 3 AUC statistics of MRSL with different MRAugment augmentation levels and MRSSL with different MRAugment level combinations
| 方法 | MRAugment等级/等级组合 | 验证集AUC | 测试集AUC | ||||
|---|---|---|---|---|---|---|---|
| 平均值 | 标准差 | 中位数 | 平均值 | 标准差 | 中位数 | ||
| MRSL | 0.936 | 0.008 | 0.938 | 0.856 | 0.012 | 0.858 | |
| 0.956 | 0.010 | 0.960 | 0.916 | 0.012 | 0.914 | ||
| 0.965 | 0.009 | 0.965 | 0.919 | 0.013 | 0.915 | ||
| 0.961 | 0.013 | 0.964 | 0.906 | 0.013 | 0.909 | ||
| MRSSL | 0.947 | 0.010 | 0.946 | 0.928 | 0.012 | 0.926 | |
| 0.954 | 0.009 | 0.957 | 0.939 | 0.006 | 0.939 | ||
| 0.960 | 0.008 | 0.960 | 0.946 | 0.010 | 0.949 | ||
| 0.951 | 0.005 | 0.952 | 0.944 | 0.015 | 0.944 | ||
| 0.951 | 0.011 | 0.950 | 0.942 | 0.009 | 0.945 | ||
| 0.951 | 0.011 | 0.952 | 0.942 | 0.008 | 0.941 | ||
| 0.949 | 0.007 | 0.947 | 0.933 | 0.010 | 0.931 | ||
 
																													图3 MRSL与MRSSL在不同MRAugment扩增等级与等级组合下的AUC变化
Fig. 3 AUC changes of MRSL with different MRAugment augmentation levels and MRSSL with different MRAugment level combinations
| 方法 | AUC | 准确率 | 敏感性 | 特异性 | G-mean | 
|---|---|---|---|---|---|
| 两者差值 | 0.043 | 0.025 | -0.009 | 0.160 | 0.085 | 
| MRSL | 0.908 [0.838,0.978] | 0.858 [0.785,0.910] | 0.916 [0.843,0.957] | 0.640 [0.445,0.798] | 0.766 | 
| MRSSL | 0.951 [0.913,0.989] | 0.883 [0.814,0.929] | 0.905 [0.830,0.949] | 0.800 [0.609,0.911] | 0.851 | 
表4 MRSL与MRSSL各自选取的代表模型在5个指标上的结果对比
Tab. 4 Result comparison between the selected models of MRSL and MRSSL on 5 evaluation indicators
| 方法 | AUC | 准确率 | 敏感性 | 特异性 | G-mean | 
|---|---|---|---|---|---|
| 两者差值 | 0.043 | 0.025 | -0.009 | 0.160 | 0.085 | 
| MRSL | 0.908 [0.838,0.978] | 0.858 [0.785,0.910] | 0.916 [0.843,0.957] | 0.640 [0.445,0.798] | 0.766 | 
| MRSSL | 0.951 [0.913,0.989] | 0.883 [0.814,0.929] | 0.905 [0.830,0.949] | 0.800 [0.609,0.911] | 0.851 | 
| 方法 | AUC | 准确率 | 敏感性 | 特异性 | G-mean | 
|---|---|---|---|---|---|
| MRSL(数据扩增) | 0.919±0.013 | 0.876±0.009 | 0.716±0.047 | 0.810±0.026 | |
| MRSL(高斯噪声) | 0.887±0.019 | 0.849±0.022 | 0.905±0.029 | 0.636±0.081 | 0.757±0.046 | 
| PI Model | 0.890±0.023 | 0.844±0.018 | 0.882±0.034 | 0.698±0.133 | 0.780±0.062 | 
| FixMatch | 0.930±0.011 | 0.875±0.021 | 0.901±0.035 | 0.778±0.078 | |
| UDA | 0.916±0.024 | 0.756±0.061 | 0.831±0.030 | ||
| MRSSL | 0.946±0.010 | 0.893±0.013 | 0.925±0.019 | 0.843±0.025 | 
表5 不同方法的结果对比
Tab. 5 Result comparison of different methods
| 方法 | AUC | 准确率 | 敏感性 | 特异性 | G-mean | 
|---|---|---|---|---|---|
| MRSL(数据扩增) | 0.919±0.013 | 0.876±0.009 | 0.716±0.047 | 0.810±0.026 | |
| MRSL(高斯噪声) | 0.887±0.019 | 0.849±0.022 | 0.905±0.029 | 0.636±0.081 | 0.757±0.046 | 
| PI Model | 0.890±0.023 | 0.844±0.018 | 0.882±0.034 | 0.698±0.133 | 0.780±0.062 | 
| FixMatch | 0.930±0.011 | 0.875±0.021 | 0.901±0.035 | 0.778±0.078 | |
| UDA | 0.916±0.024 | 0.756±0.061 | 0.831±0.030 | ||
| MRSSL | 0.946±0.010 | 0.893±0.013 | 0.925±0.019 | 0.843±0.025 | 
| 1 | LITJENS G, KOOI T, EHTESHAMI BEJNORDI B, et al. A survey on deep learning in medical image analysis[J]. Medical Image Analysis, 2017, 42:60-88. 10.1016/j.media.2017.07.005 | 
| 2 | SHEN D G, WU G R, SUK H I. Deep learning in medical image analysis[J]. Annual Review of Biomedical Engineering, 2017, 19: 221-248. 10.1146/annurev-bioeng-071516-044442 | 
| 3 | SONG Y, ZHANG Y D, YAN X, et al. Computer-aided diagnosis of prostate cancer using a deep convolutional neural network from multiparametric MRI[J]. Journal of Magnetic Resonance Imaging, 2018, 48(6): 1570-1577. 10.1002/jmri.26047 | 
| 4 | HERENT P, SCHMAUCH B, JEHANNO P, et al. Detection and characterization of MRI breast lesions using deep learning[J]. Diagnostic and Interventional Imaging, 2019, 100(4): 219-225. 10.1016/j.diii.2019.02.008 | 
| 5 | ZHOU Z Y, ZHAO G Y, KIJOWSKI R, et al. Deep convolutional neural network for segmentation of knee joint anatomy[J]. Magnetic Resonance in Medicine, 2018, 80(6): 2759-2770. 10.1002/mrm.27229 | 
| 6 | HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 770-778. 10.1109/cvpr.2016.90 | 
| 7 | HE K M, ZHANG X Y, REN S Q, et al. Identity mappings in deep residual networks[C]// Proceedings of the 2016 European Conference on Computer Vision, LNCS9908. Cham: Springer, 2016: 630-645. | 
| 8 | ZAGORUYKO S, KOMODAKIS N. Wide residual networks[C]// Proceedings of the 2016 British Machine Vision Conference. Durham: BMVA Press, 2016: No.87. 10.5244/c.30.87 | 
| 9 | PAN S J, YANG Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359. 10.1109/tkde.2009.191 | 
| 10 | BIEN N, RAJPURKAR P, BALL RL, et al. Deep-learning-assisted diagnosis for knee magnetic resonance imaging: development and retrospective validation of MRNet[J]. PLoS Medicine, 2018, 15(11): No.e1002699. 10.1371/journal.pmed.1002699 | 
| 11 | DENG J, DONG W, SOCHER R, et al. ImageNet: a large-scale hierarchical image database[C]// Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2009: 248-255. 10.1109/cvpr.2009.5206848 | 
| 12 | ZHU X J, GOLDBERG A B. Introduction to Semi-Supervised Learning[M]. San Rafael, CA: Morgan & Claypool Publishers, 2009: 23-33. 10.3115/1621829.1621832 | 
| 13 | LAINE S, AILA T. Temporal ensembling for semi-supervised learning[EB/OL]. (2017-03-15) [2020-12-12].. | 
| 14 | MIYATO T, MAEDA S I, KOYAMA M, et al. Virtual adversarial training: a regularization method for supervised and semi-supervised learning[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(8): 1979-1993. 10.1109/tpami.2018.2858821 | 
| 15 | TARVAINEN A, VALPOLA H. Mean teachers are better role models: weight-averaged consistency targets improve semi-supervised deep learning results[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY: Curran Associates, Inc., 2017: 1195-1204. | 
| 16 | XIE Q Z, DAI Z H, HOVY E, et al. Unsupervised data augmentation for consistency training[C/OL]// Proceedings of the 34th International Conference on Neural Information Processing Systems. [2020-12-12].. | 
| 17 | CHEPLYGINA V, DE BRUIJNE M, PLUIM J P W. Not-so-supervised: a survey of semi-supervised, multi-instance, and transfer learning in medical image analysis[J]. Medical Image Analysis, 2019, 54: 280-296. 10.1016/j.media.2019.03.009 | 
| 18 | LIU Q D, YU L Q, LUO L Y, et al. Semi-supervised medical image classification with relation-driven self-ensembling model[J]. IEEE Transactions on Medical Imaging, 2020, 39(11): 3429-3440. 10.1109/tmi.2020.2995518 | 
| 19 | NGUYEN H H, SAARAKKALA S, BLASCHKO M B, et al. Semixup: in- and out-of-manifold regularization for deep semi-supervised knee osteoarthritis severity grading from plain radiographs[J]. IEEE Transactions on Medical Imaging, 2020, 39: 4346-4356. 10.1109/tmi.2020.3017007 | 
| 20 | PETERFY C G, SCHNEIDER E, NEVITT M. The osteoarthritis initiative: report on the design rationale for the magnetic resonance imaging protocol for the knee[J]. Osteoarthritis and Cartilage, 2008, 16(12): 1433-1441. 10.1016/j.joca.2008.06.016 | 
| 21 | HU J L, KUANG Y Z, LIAO B, et al. A multichannel 2D convolutional neural network model for task-evoked fMRI data classification[J]. Computational Intelligence and Neuroscience, 2019, 2019: No.5065214. 10.1155/2019/5065214 | 
| 22 | MITCHELL T M. The need for biases in learning generalizations: Rutgers CS tech report CBM-TR-117[R]. New Brunswick, NJ: Rutgers University, 1980: 1-3. | 
| 23 | CUBUK E D, ZOPH B, SHLENS J, et al. RandAugment: practical automated data augmentation with a reduced search space[C]// Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Piscataway: IEEE, 2020: 3008-3017. 10.1109/cvprw50498.2020.00359 | 
| 24 | NYÚL L G, UDUPA J K. On standardizing the MR image intensity scale[J]. Magnetic Resonance in Medicine, 1999, 42(6): 1072-1081. 10.1002/(sici)1522-2594(199912)42:6<1072::aid-mrm11>3.0.co;2-m | 
| 25 | LOWEKAMP B C, CHEN D T, IBÁÑEZ L, et al. The design of simpleITK[J]. Frontiers in Neuroinformatics, 2013, 7: No.45. 10.3389/fninf.2013.00045 | 
| 26 | YANIV Z, LOWEKAMP B C, JOHNSON H J, et al. SimpleITK image-analysis notebooks: a collaborative environment for education and reproducible research[J]. Journal of Digital Imaging, 2018, 31(3): 290-303. 10.1007/s10278-017-0037-8 | 
| 27 | WALT S VAN DER, SCHÖNBERGER J L, NUNEZ-IGLESIAS J, et al. Scikit-image: image processing in Python[J]. PeerJ, 2014, 2: No.e453. 10.7717/peerj.453 | 
| 28 | JAPKOWICZ N, STEPHEN S. The class imbalance problem: a systematic study[J]. Intelligent Data Analysis, 2002, 6(5): 429-449. 10.3233/ida-2002-6504 | 
| 29 | SOHN K, BERTHELOT D, LI C L, et al. FixMatch: simplifying semi-supervised learning with consistency and confidence[C/OL]// Proceedings of the 34th International Conference on Neural Information Processing Systems. [2020-12-12].. | 
| 30 | PASZKE A, GROSS S, MASSA F, et al. PyTorch: an imperative style, high-performance deep learning library[C/OL]// Proceedings of the 33rd International Conference on Neural Information Processing Systems. [2020-12-12].. | 
| [1] | 秦璟, 秦志光, 李发礼, 彭悦恒. 基于概率稀疏自注意力神经网络的重性抑郁疾患诊断[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2970-2974. | 
| [2] | 王熙源, 张战成, 徐少康, 张宝成, 罗晓清, 胡伏原. 面向手术导航3D/2D配准的无监督跨域迁移网络[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2911-2918. | 
| [3] | 黄云川, 江永全, 黄骏涛, 杨燕. 基于元图同构网络的分子毒性预测[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2964-2969. | 
| [4] | 李顺勇, 李师毅, 胥瑞, 赵兴旺. 基于自注意力融合的不完整多视图聚类算法[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2696-2703. | 
| [5] | 潘烨新, 杨哲. 基于多级特征双向融合的小目标检测优化模型[J]. 《计算机应用》唯一官方网站, 2024, 44(9): 2871-2877. | 
| [6] | 张英俊, 李牛牛, 谢斌红, 张睿, 陆望东. 课程学习指导下的半监督目标检测框架[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2326-2333. | 
| [7] | 刘禹含, 吉根林, 张红苹. 基于骨架图与混合注意力的视频行人异常检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2551-2557. | 
| [8] | 顾焰杰, 张英俊, 刘晓倩, 周围, 孙威. 基于时空多图融合的交通流量预测[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2618-2625. | 
| [9] | 石乾宏, 杨燕, 江永全, 欧阳小草, 范武波, 陈强, 姜涛, 李媛. 面向空气质量预测的多粒度突变拟合网络[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2643-2650. | 
| [10] | 吴筝, 程志友, 汪真天, 汪传建, 王胜, 许辉. 基于深度学习的患者麻醉复苏过程中的头部运动幅度分类方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2258-2263. | 
| [11] | 李欢欢, 黄添强, 丁雪梅, 罗海峰, 黄丽清. 基于多尺度时空图卷积网络的交通出行需求预测[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2065-2072. | 
| [12] | 张郅, 李欣, 叶乃夫, 胡凯茜. 基于暗知识保护的模型窃取防御技术DKP[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2080-2086. | 
| [13] | 赵亦群, 张志禹, 董雪. 基于密集残差物理信息神经网络的各向异性旅行时计算方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2310-2318. | 
| [14] | 徐松, 张文博, 王一帆. 基于时空信息的轻量视频显著性目标检测网络[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2192-2199. | 
| [15] | 孙逊, 冯睿锋, 陈彦如. 基于深度与实例分割融合的单目3D目标检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2208-2215. | 
| 阅读次数 | ||||||
| 全文 |  | |||||
| 摘要 |  | |||||