Application of image segmentation methods based on deep learning in aortic diseases
-
摘要: 主动脉疾病(如主动脉瘤、主动脉夹层等)严重危害患者生命健康，若患者得不到及时治疗，其后果通常是致命的。借助图像分割技术精准识别出患者病灶区域，医生可以实现对疾病的精确诊断、术前规划或术后监控等。最近深度学习在医学图像分割任务中展现出明显优势，越来越多的学者将其应用于主动脉疾病领域。本研究旨在对基于深度学习的图像分割方法在主动脉疾病中的应用进行综述。Abstract: Aortic diseases, such as aortic aneurysm, aortic dissection, etc., seriously endanger patients'health. If patients are not treated in time, the consequences are usually fatal. With image segmentation to accurately identify patients'lesion area, doctors can achieve an accurate diagnosis, preoperative planning or postoperative monitoring. Recently, deep learning has shown obvious advantages in medical image segmentation tasks. An increasing number of scholars apply deep learning to aortic diseases. This paper aims to review the application of image segmentation methods based on deep learning in aortic diseases.
-
Key words:
- deep learning /
- image segmentation /
- aortic diseases
-
-
表 1 主动脉分割的代表性研究
Table 1. Summary of representative studies on aortic segmentation
文献 影像来源 数据集 分割区域 网络介绍 分割结果 Baskaran等[16] 冠状动脉CTA 206例患者(Tr=144，V=42，Te=20) 近端升主动脉降主动脉 改进的U-Net DSC中位数分别为0.97和0.95 Sieren等[17] CTA 191例患者(Tr：V：Te=7：1：2) 主动脉、AA、AD 3D U-Net DSC中位数为0.95，HSD中位数为8 mm Fantazzini等[18] CTA增强扫描 80例患者(Tr=64，V=6，Te=10) 主动脉及分支血管 3个正交的2D U-Net DSC为0.92±0.01 吉淑滢等[19] 胸部CT图像 ISBI 2019数据集(Tr=40，Te=20) 胸主动脉 改进的U-Net DSC为0.94，HSD为0.19 mm 向曦[20] 心脏CT 87例患者 胸主动脉 改进的V-Net DSC为0.90，IoU为0.81 注：Tr指训练集，V指验证集，Te指测试集。 
下载: 导出CSV
表 2 AA的代表性研究
Table 2. Summary of representative studies on AA
文献 影像来源 数据集 分割区域 网络介绍 分割结果 Comelli等[22] 心电图门控CTA 72例患者(Tr：Te=4：1) 升胸主动脉瘤 UNet、ENet和ERFNet 三者DSC均高于0.88 Adam等[23] CTA 551个数据集(Tr=489，V=62) AA(自动直径测量) V-Net架构 健康、疾病和术后：平均DSC分别为0.84、0.95和0.93 Hepp等[24] MRI 100个数据集(Tr=70，Te=30) TAA(自动直径测量) 深度CNN 平均DSC为0.85 Bratt等[25] CTA 2835例患者 TAA(自动直径测量) 改进的U-Net 平均DSC为0.96 Saitta等[26] 3D CT 465例患者(Tr=395，Te=70) TAA(自动直径测量) 3D U-Net 平均DSC为0.95 Caradu等[14] CTA 100例患者 AAA、腔内血栓(自动直径测量) 类似U-Net的CNN网络 DSC为0.95±0.01，HSD为(4.61±7.26) mm Lu等[27] CT、CTA 378个数据集(Tr+V=321，Te=57) AAA(自动直径测量) 3D U-Net 平均DSC为0.91 Brutti等[28] CTA 85例患者(Tr=63，V=8，Te=14) 腔内血栓 U-Net 平均DSC为0.89 Lareyre等[29] CTA 93例患者 腹主动脉、血栓 U-Net和专家系统混合模型 DSC中位数分别为0.92、0.82 Duo等[30] CT，MRI 21例患者(Tr：V：Te=8：1：1) 主动脉、血栓和钙化 CNN融合模型 准确率超过98%
下载: 导出CSV
表 3 AD的代表性研究
Table 3. Summary of representative studies on AD
文献 影像来源 数据集 分割区域 网络介绍 分割结果 Cheng等[31] CT 20例患者(Tr：V：Te=3：1：1) 主动脉分割与夹层检测 改进的U-Net DSC为0.91，IoU为0.95 Li等[32] CT 45例患者 主动脉、真腔 级联卷积网络 平均DSC为0.99、0.93 Lyu等[33] CTA 42例患者(Tr：Te=5：1) AD 3D CNN 平均DSC为0.92 Cao等[34] 术前CTA 276例患者(Tr=246，Te=30) TBAD的主动脉、真腔、假腔 CNN变体 最佳平均DSC为0.93、0.93、0.91 Chen等[35] CTA 120例患者 TBAD真腔、假腔、分支血管 级联神经网络 平均DSC为0.96、0.95、0.89 Wobben等[36] CTA 40例患者(Tr=28，V=6，Te=6) TBAD真腔、假腔、假腔内血栓 3D residual U-Net DSC中位数分别为0.92、0.91、0.78 Yu等[37] CTA 139例患者(Tr=99，V=15，Te=25) TBAD的主动脉、真腔、假腔 三维深度CNN 平均DSC分别为0.96、0.96、0.93
下载: 导出CSV
-
[1] Bossone E, Eagle KA. Epidemiology and management of aortic disease: aortic aneurysms and acute aortic syndromes[J]. Nat Rev Cardiol, 2021, 18(5): 331-348. doi: 10.1038/s41569-020-00472-6
[2] 蔡治祥, 颜涛, 王显悦, 等. 非编码RNA与获得性主动脉疾病的研究进展[J]. 临床心血管病杂志, 2021, 37(4): 370-374. https://www.cnki.com.cn/Article/CJFDTOTAL-LCXB202104016.htm
[3] 曹玉红, 徐海, 刘荪傲, 等. 基于深度学习的医学影像分割研究综述[J]. 计算机应用, 2021, 41(8): 2273-2287. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJY202108015.htm
[4] Hahn LD, Baeumler K, Hsiao A. Artificial intelligence and machine learning in aortic disease[J]. Curr Opin Cardiol, 2021, 36(6): 695-703. doi: 10.1097/HCO.0000000000000903
[5] Zhou T, Ruan S, Canu S. A review: Deep learning for medical image segmentation using multi-modality fusion[J]. Array, 2019, 3-4: 100004. doi: 10.1016/j.array.2019.100004
[6] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]//Conference on Computer Vision and Pattern Recognition. 2015 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Boston, MA, USA: IEEE, 2015: 3431-3440.
[7] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation[C]//Navab N, Hornegger J, Wells W, et al. MICCAI 2015: Medical Image Computing and Computer-Assisted Intervention-MICCAI 2015, vol 9351. Munich, Germany: Springer, Cham, 2015: 234-241.
[8] Zhou Z, Siddiquee M, Tajbakhsh N, et al. UNet++: A Nested U-Net Architecture for Medical Image Segmentation[J]. Deep Learn Med Image Anal Multimodal Learn Clin Decis Support(2018), 2018, 11045: 3-11.
[9] Oktay O, Schlemper J, Le Folgoc L, et al. Attention U-Net: Learning Where to Look for the Pancreas[J/OL]. arXiv: 1804.03999, 2018.
[10] Jha D, Smedsrud P H, Riegler M A, et al. ResUNet++: An Advanced Architecture for Medical Image Segmentation[C]//Bulterman D, Kankanhalli M, Muehlhaeuser M, et al. 2019 IEEE International Symposium on Multimedia(ISM). San Diego, CA, USA: IEEE, 2019: 225-230.
[11] Çiçek Ö, Abdulkadir A, Lienkamp SS, et al. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation[C]//Ourselin S, Joskowicz L, Sabuncu M, et al. MICCAI 2016: Medical Image Computing and Computer-Assisted Intervention-MICCAI 2016, vol 9901. Athens, Greece: Springer, Cham, 2016: 424-432.
[12] Milletari F, Navab N, Ahmadi S. V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation[C]//Savarese S. 2016 Fourth International Conference on 3D Vision(3DV). Stanford, CA, USA: IEEE, 2016: 565-571.
[13] Isensee F, Jaeger PF, Kohl S, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation[J]. Nat Methods, 2021, 18(2): 203-211. doi: 10.1038/s41592-020-01008-z
[14] Caradu C, Spampinato B, Vrancianu AM, et al. Fully automatic volume segmentation of infrarenal abdominal aortic aneurysm computed tomography images with deep learning approaches versus physician controlled manual segmentation[J]. J Vasc Surg, 2021, 74(1): 246-256. e6. doi: 10.1016/j.jvs.2020.11.036
[15] Moccia S, De Momi E, El Hadji S, et al. Blood vessel segmentation algorithms-Review of methods, datasets and evaluation metrics[J]. Comput Methods Programs Biomed, 2018, 158: 71-91. doi: 10.1016/j.cmpb.2018.02.001
[16] Baskaran L, Al'Aref SJ, Maliakal G, et al. Automatic segmentation of multiple cardiovascular structures from cardiac computed tomography angiography images using deep learning[J]. PLoS One, 2020, 15(5): e0232573. doi: 10.1371/journal.pone.0232573
[17] Sieren MM, Widmann C, Weiss N, et al. Automated segmentation and quantification of the healthy and diseased aorta in CT angiographies using a dedicated deep learning approach[J]. Eur Radiol, 2022, 32(1): 690-701. doi: 10.1007/s00330-021-08130-2
[18] Fantazzini A, Esposito M, Finotello A, et al. 3D Automatic Segmentation of Aortic Computed Tomography Angiography Combining Multi-View 2D Convolutional Neural Networks[J]. Cardiovasc Eng Technol, 2020, 11(5): 576-586. doi: 10.1007/s13239-020-00481-z
[19] 吉淑滢, 肖志勇. 融合上下文和多尺度特征的胸部多器官分割[J]. 中国图象图形学报, 2021, 26(9): 2135-2145. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB202109008.htm
[20] 向曦. 基于深度学习的3D精细化主动脉管状结构分割方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2021.
[21] Pape LA, Tsai TT, Isselbacher EM, et al. Aortic Diameter ≥5.5 cm Is Not a Good Predictor of Type A Aortic Dissection[J]. Circulation, 2007, 116(10): 1120-1127. doi: 10.1161/CIRCULATIONAHA.107.702720
[22] Comelli A, Dahiya N, Stefano A, et al. Deep learning approach for the segmentation of aneurysmal ascending aorta[J]. Biomed Eng Lett, 2021, 11(1): 15-24. doi: 10.1007/s13534-020-00179-0
[23] Adam C, Fabre D, Mougin J, et al. Pre-surgical and Post-surgical Aortic Aneurysm Maximum Diameter Measurement: Full Automation by Artificial Intelligence[J]. Eur J Vasc Endovasc Surg, 2021, 62(6): 869-877. doi: 10.1016/j.ejvs.2021.07.013
[24] Hepp T, Fischer M, Winkelmann MT, et al. Fully Automated Segmentation and Shape Analysis of the Thoracic Aorta in Non-contrast-enhanced Magnetic Resonance Images of the German National Cohort Study[J]. J Thorac Imaging, 2020, 35(6): 389-398.
[25] Bratt A, Blezek DJ, Ryan WJ, et al. Deep Learning Improves the Temporal Reproducibility of Aortic Measurement[J]. J Digit Imaging, 2021, 34(5): 1183-1189. doi: 10.1007/s10278-021-00465-y
[26] Saitta S, Sturla F, Caimi A, et al. A Deep Learning-Based and Fully Automated Pipeline for Thoracic Aorta Geometric Analysis and Planning for Endovascular Repair from Computed Tomography[J]. J Digit Imaging, 2022, 35(2): 226-239. doi: 10.1007/s10278-021-00535-1
[27] Lu JT, Brooks R, Hahn S, et al. DeepAAA: Clinically Applicable and Generalizable Detection of Abdominal Aortic Aneurysm Using Deep Learning[C]//Shen D, Liu T, Peters T M, et al. MICCAI 2019: Medical Image Computing and Computer Assisted Intervention-MICCAI 2019, vol 11765. Shenzhen, China: Springer, Cham, 2019: 723-731.
[28] Brutti F, Fantazzini A, Finotello A, et al. Deep Learning to Automatically Segment and Analyze Abdominal Aortic Aneurysm from Computed Tomography Angiography[J]. Cardiovasc Eng Technol, 2022.
[29] Lareyre F, Adam C, Carrier M, et al. Automated Segmentation of the Human Abdominal Vascular System Using a Hybrid Approach Combining Expert System and Supervised Deep Learning[J]. J Clin Med, 2021, 10(15).
[30] Duo W, Rui Z, Jin Z, et al. Neural network fusion: a novel CT-MR aortic aneurysm image segmentation method[C]//Angelini E D, Landman B A. Proc. SPIE 10574, Medical Imaging 2018: Image Processing. Houston, Texas, United States: SPIE, 2018: 1057424.
[31] Cheng J, Tian S, Yu L, et al. A deep learning algorithm using contrast-enhanced computed tomography(CT)images for segmentation and rapid automatic detection of aortic dissection[J]. Biomed Signal Process Control, 2020, 62: 102145. doi: 10.1016/j.bspc.2020.102145
[32] Li Z, Feng J, Feng Z, et al. Lumen Segmentation of Aortic Dissection with Cascaded Convolutional Network[C]//Pop M, Sermesant M, Zhao J, et al. STACOM 2018: Statistical Atlases and Computational Models of the Heart Atrial Segmentation and LV Quantification Challenges, vol 11395. Granada, Spain: Springer, Cham, 2019: 122-130.
[33] Lyu T, Yang G, Zhao X, et al. Dissected aorta segmentation using convolutional neural networks[J]. Comput Methods Programs Biomed, 2021, 211: 106417. doi: 10.1016/j.cmpb.2021.106417
[34] Cao L, Shi R, Ge Y, et al. Fully automatic segmentation of type B aortic dissection from CTA images enabled by deep learning[J]. Eur J Radiol, 2019, 121: 108713. doi: 10.1016/j.ejrad.2019.108713
[35] Chen D, Zhang X, Mei Y, et al. Multi-stage learning for segmentation of aortic dissections using a prior aortic anatomy simplification[J]. Med Image Anal, 2021, 69: 101931. doi: 10.1016/j.media.2020.101931
[36] Wobben LD, Codari M, Mistelbauer G, et al. Deep Learning-Based 3D Segmentation of True Lumen, False Lumen, and False Lumen Thrombosis in Type-B Aortic Dissection[J]. Annu Int Conf IEEE Eng Med Biol Soc, 2021, 2021: 3912-3915.
[37] Yu Y, Gao Y, Wei J, et al. A Three-Dimensional Deep Convolutional Neural Network for Automatic Segmentation and Diameter Measurement of Type B Aortic Dissection[J]. Korean J Radiol, 2021, 22(2): 168-178. doi: 10.3348/kjr.2020.0313
[38] Stather PW, Sidloff DA, Dattani N, et al. Authors' reply: Systematic review and meta-analysis of the early and late outcomes of open and endovascular repair of abdominal aortic aneurysm[J]. Br J Surg, 2013, 100(11): 863-872.
[39] Hahn S, Perry M, Morris CS, et al. Machine deep learning accurately detects endoleak after endovascular abdominal aortic aneurysm repair[J]. JVS Vasc Sci, 2020, 1: 5-12. doi: 10.1016/j.jvssci.2019.12.003
[40] López-Linares K, Aranjuelo N, Kabongo L, et al. Fully automatic detection and segmentation of abdominal aortic thrombus in post-operative CTA images using Deep Convolutional Neural Networks[J]. Med Image Anal, 2018, 46: 202-214. doi: 10.1016/j.media.2018.03.010
[41] Talebi S, Madani MH, Madani A, et al. Machine learning for endoleak detection after endovascular aortic repair[J]. Sci Rep, 2020, 10(1): 18343. doi: 10.1038/s41598-020-74936-7
-