1172. 工程科学学报，第43卷.第9期 (a) P(ASDF36.0%,(b) P(ASD=97.2% (陶攀，付忠良，朱错，等.基于深度学习的超声心动图切面识别 方法.计算机应用，2017,37(5)：1434) [5]Madani A,Amaout R,Mofrad M,et al.Fast and accurate view classification of echocardiograms using deep learning.Npi Digit Med,2018,1(1):6 [61 Madani A,Ong J R,Tibrewal A,et al.Deep echocardiography: Data-efficient supervised and semi-supervised deep learning (c) P(ASD=25.1% (d) P(ASD=88.1% towards automated diagnosis of cardiac disease.Npj Digit Med, 2018,1(1:59 [7] Teng L,Fu Z L,Yao Y.Interactive translation in echocardiography training system with enhanced cycle-GAN. IEEE Access3,2020,8:106147 图7房间隔遮挡测试(a,c)遮挡前：(b,d)遮挡后 [8]Teng L,Fu Z L,Ma Q,et al.Interactive echocardiography translation using few-shot GAN transfer learning.Comput Math Fig.7 Atrial septal covering test:(a,c)before covering:(b,d)covered Methods Med.2020,2020:1487035 4结论 [9]Ghorbani A,Ouyang D,Abid A,et al.Deep learning interpretation of echocardiograms.Npj Digit Med,2020,3(1):1 提出了一种基于超声心动图特征切面识别的 [10]Veni G,Moradi M,Bulu H K,et al.Echocardiography ASD辅助诊断模型，并通过实验测试验证了所提 segmentation based on a shape-guided deformable model driven 模型取得了非常不错的诊断性能.同时应用双边 by a fully convolutional network prior /2018 IEEE 15th 滤波算法对超声心动图进行噪声抑制，同时保留 International Symposium on Biomedical Imaging (ISBI 2018). 轮廓边界，使得模型诊断性能进一步提升.再通过 Washington,2018:898 房间隔处的遮挡测试，验证了所提模型做出 [11]Leclerc S,Smistad E,Pedrosa J,et al.Deep learning for ASD诊断的依据与标准与临床诊断保持一致，从 segmentation using an open large-scale dataset in 2D echocardiography.IEEE Trans Med Imaging,2019,38(9):2198 而说明了其诊断结果的可靠性.本文给出了一种 [12]Li YW,HoCP,Toulemonde M,et al.Fully automatic myocardial 针对有着不同特征的医学影像数据进行疾病诊断 segmentation of contrast echocardiography sequence using random 的可行思路，对未来基于深度学习的超声心动图 forests guided by shape model.IEEE Trans Med Imaging.2017. 相关的疾病诊断模型设计具有一定的参考价值， 37(5):1081 而将超声检查报告、病人病历等多模态数据融入 [13]Lu Y,Fu X H,Li X Q,et al.Cardiac chamber segmentation using 到其中也可能是进一步提高诊断性能的可行思 deep learning on magnetic resonance images from patients before 路，在此留给未来的研究者进行探索 and after atrial septal occlusion surgery //2020 42nd Anmual International Conference of the IEEE Engineering in Medicine Biology Society (EMBC).Montreal,2020:1211 参考文献 [14]Zyuzin V,Mukhtarov A,Neustroev D,et al.Segmentation of 2D [1]Wang X F,Xie M X.Textbook of Echocardiograply.5th Ed. echocardiography images using residual blocks in U-net Beijing:People's Medical Publishing House,2016 architectures /2020 Ural Symposium on Biomedical Engineering. (王新房，谢明星.超声心动图学.5版.北京：人民卫生出版社， Radioelectronics and Information Technology (USBEREIT). 2016) Yekaterinburg,2020:499 [2]Li J W.Trend and epidemiological analysis of congenital heart [15]Smistad E,Ostvik A,Salte I M,et al.Fully automatic real-time disease.Chin J Cardiovasc Rehabilitation Med.2017.26(1):60 ejection fraction and MAPSE measurements in 2D (黎洁雯.先天性心脏病的流行趋势及流行病学分析.心血管康 echocardiography using deep neural networks //2018 IEEE 复医学杂志，2017,26(1)：60) International Uhrasonics Symposium (IUS).Kobe,01:1 [3]Zhang L Y,Qiao Y H,Ning S F,et al.Analysis of ultrasonic [16]Davis A,Billick K,Horton K,et al.Artificial intelligence and diagnosis of 39 atrial septal defects.China Pract Med,2009, echocardiography:A primer for cardiac sonographers.m Soc 4(14):95 Echocardiogr,2020,33(9):1061 (张丽媛，乔玉红，宁淑范，等.房间隔缺损39例超声诊断分析 [17]Huang Q.Zhang F,Li X.Machine learning in ultrasound 中国实用医药，2009,4(14)：95) computer-aided diagnostic systems:A survey.Biomed Res Int, [4]Tao P,Fu Z L,Zhu K,et al.Echocardiogram view recognition 2018,2018:5137904 using deep convolutional neural network.J Compur Appl,2017, [18]Wang X,Jia Y G,Sevenster M,et al.Representation learning of 37(5):1434 finding codes in structured echocardiogram reporting //2018 /EEE4    结论 提出了一种基于超声心动图特征切面识别的 ASD 辅助诊断模型，并通过实验测试验证了所提 模型取得了非常不错的诊断性能. 同时应用双边 滤波算法对超声心动图进行噪声抑制，同时保留 轮廓边界，使得模型诊断性能进一步提升. 再通过 房 间 隔 处 的 遮 挡 测 试 ， 验 证 了 所 提 模 型 做 出 ASD 诊断的依据与标准与临床诊断保持一致，从 而说明了其诊断结果的可靠性. 本文给出了一种 针对有着不同特征的医学影像数据进行疾病诊断 的可行思路，对未来基于深度学习的超声心动图 相关的疾病诊断模型设计具有一定的参考价值， 而将超声检查报告、病人病历等多模态数据融入 到其中也可能是进一步提高诊断性能的可行思 路，在此留给未来的研究者进行探索. 参    考    文    献 Wang  X  F,  Xie  M  X. Textbook of Echocardiography.  5th  Ed. Beijing: People’s Medical Publishing House, 2016 （ 王新房, 谢明星. 超声心动图学. 5版. 北京: 人民卫生出版社, 2016） [1] Li  J  W.  Trend  and  epidemiological  analysis  of  congenital  heart disease. Chin J Cardiovasc Rehabilitation Med, 2017, 26（1）: 60 （黎洁雯. 先天性心脏病的流行趋势及流行病学分析. 心血管康 复医学杂志, 2017, 26（1）：60） [2] Zhang  L  Y,  Qiao  Y  H,  Ning  S  F,  et  al.  Analysis  of  ultrasonic diagnosis  of  39  atrial  septal  defects. China Pract Med,  2009, 4（14）: 95 （张丽媛, 乔玉红, 宁淑范, 等. 房间隔缺损39例超声诊断分析. 中国实用医药, 2009, 4（14）：95） [3] Tao  P,  Fu  Z  L,  Zhu  K,  et  al.  Echocardiogram  view  recognition using  deep  convolutional  neural  network. J Comput Appl,  2017, 37（5）: 1434 [4] （陶攀, 付忠良, 朱锴, 等. 基于深度学习的超声心动图切面识别 方法. 计算机应用, 2017, 37（5）：1434） Madani  A,  Arnaout  R,  Mofrad  M,  et  al.  Fast  and  accurate  view classification  of  echocardiograms  using  deep  learning. Npj Digit Med, 2018, 1（1）: 6 [5] Madani  A,  Ong  J  R,  Tibrewal  A,  et  al.  Deep  echocardiography: Data-efficient  supervised  and  semi-supervised  deep  learning towards  automated  diagnosis  of  cardiac  disease. Npj Digit Med, 2018, 1（1）: 59 [6] Teng  L,  Fu  Z  L,  Yao  Y.  Interactive  translation  in echocardiography  training  system  with  enhanced  cycle-GAN. IEEE Access, 2020, 8: 106147 [7] Teng  L,  Fu  Z  L,  Ma  Q,  et  al.  Interactive  echocardiography translation  using  few-shot  GAN  transfer  learning. Comput Math Methods Med, 2020, 2020: 1487035 [8] Ghorbani A, Ouyang D, Abid A, et al. Deep learning interpretation of echocardiograms. Npj Digit Med, 2020, 3（1）: 1 [9] Veni  G,  Moradi  M,  Bulu  H  K,  et  al.  Echocardiography segmentation  based  on  a  shape-guided  deformable  model  driven by  a  fully  convolutional  network  prior  //  2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018). Washington, 2018: 898 [10] Leclerc  S,  Smistad  E,  Pedrosa  J,  et  al.  Deep  learning  for segmentation  using  an  open  large-scale  dataset  in  2D echocardiography. IEEE Trans Med Imaging, 2019, 38（9）: 2198 [11] Li Y W, Ho C P, Toulemonde M, et al. Fully automatic myocardial segmentation of contrast echocardiography sequence using random forests  guided  by  shape  model. IEEE Trans Med Imaging,  2017, 37（5）: 1081 [12] Lu Y, Fu X H, Li X Q, et al. Cardiac chamber segmentation using deep learning on magnetic resonance images from patients before and  after  atrial  septal  occlusion  surgery  //  2020  42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). Montreal, 2020: 1211 [13] Zyuzin V, Mukhtarov A, Neustroev D, et al. Segmentation of 2D echocardiography  images  using  residual  blocks  in  U-net architectures // 2020 Ural Symposium on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT). Yekaterinburg, 2020: 499 [14] Smistad  E,  Østvik  A,  Salte  I  M,  et  al.  Fully  automatic  real-time ejection  fraction  and  MAPSE  measurements  in  2D echocardiography  using  deep  neural  networks  //  2018 IEEE International Ultrasonics Symposium (IUS). Kobe, 2018: 1 [15] Davis  A,  Billick  K,  Horton  K,  et  al.  Artificial  intelligence  and echocardiography:  A  primer  for  cardiac  sonographers. J Am Soc Echocardiogr, 2020, 33（9）: 1061 [16] Huang  Q,  Zhang  F,  Li  X.  Machine  learning  in  ultrasound computer-aided  diagnostic  systems:  A  survey. Biomed Res Int, 2018, 2018: 5137904 [17] Wang X, Jia Y G, Sevenster M, et al. Representation learning of finding codes in structured echocardiogram reporting // 2018 IEEE [18] (a) P (ASD)=36.0% P (ASD)=97.2% P (ASD)=25.1% P (ASD)=88.1% (b) (c) (d) 图 7    房间隔遮挡测试. (a，c) 遮挡前；(b，d) 遮挡后 Fig.7    Atrial septal covering test: (a,c) before covering; (b,d)covered · 1172 · 工程科学学报，第 43 卷，第 9 期