基于模糊标签和深度学习的TOF-MRA影像脑动脉瘤自动检测

doi:10.11938/cjmr20223004

波谱学杂志 ›› 2022, Vol. 39 ›› Issue (3): 267-277.doi: 10.11938/cjmr20223004

基于模糊标签和深度学习的TOF-MRA影像脑动脉瘤自动检测

陈萌¹,耿辰²,李郁欣³,耿道颖³,鲍奕仿^3,*(),戴亚康^1,^2,*()

1. 徐州医科大学医学影像学院, 江苏徐州 221000
2. 中国科学院苏州生物医学工程技术研究所, 江苏苏州 215000
3. 复旦大学附属华山医院放射科, 复旦大学医学功能与分子影像研究所, 上海 200000

收稿日期:2022-05-24 出版日期:2022-09-05 发布日期:2022-07-20
通讯作者: 鲍奕仿,戴亚康 E-mail:bao_yifang@163.com;daiyk@sibet.ac.cn
基金资助:
国家自然科学基金资助项目(61672236);上海市科学技术委员会科技创新行动计划临床医学领域项目(19411951200);苏州市科技发展计划项目(SS202072)

Automatic Detection for Cerebral Aneurysms in TOF-MRA Images Based on Fuzzy Label and Deep Learning

Meng CHEN¹,Chen GENG²,Yu-xin LI³,Dao-ying GENG³,Yi-fang BAO^3,*(),Ya-kang DAI^1,^2,*()

1. School of Medical Imaging, Xuzhou Medical University, Xuzhou 221000, China
2. Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou 215000, China
3. Department of Radiology, Huashan Hospital, Fudan University, Institute of Functional and Molecular Medical Imaging, Fudan University, Shanghai 200000, China

Received:2022-05-24 Online:2022-09-05 Published:2022-07-20
Contact: Yi-fang BAO,Ya-kang DAI E-mail:bao_yifang@163.com;daiyk@sibet.ac.cn

摘要/Abstract

摘要：

脑动脉瘤破裂造成的蛛网膜下腔出血致死致残率极高，借助深度学习网络辅助医生实现高效筛查具有重要意义．为提高基于时间飞跃法磁共振血管造影（Time of Flight-Magnetic Resonance Angiography，TOF-MRA）的脑动脉瘤自动检测的精度，本文基于模糊标签方式，提出一种基于变体3D U-Net和双分支通道注意力（Dual-branch Channel Attention，DCA）的深度神经网络DCAU-Net，DCA模块可以自适应地调整通道特征的响应，提高特征提取能力．首先对260例病例的TOF-MRA影像预处理，将数据集分为174例训练集、43例验证集和43例测试集，然后使用处理后的数据训练和验证DCAU-Net，测试集实验结果表明DCAU-Net可以达到90.69%的敏感度，0.83个/例的假阳性计数和0.52的阳性预测值，有望为动脉瘤筛查提供参考．

关键词: 脑动脉瘤, 自动检测, 深度学习, 模糊标签, 双分支注意力

Abstract:

Subarachnoid hemorrhage caused by the rupture of cerebral aneurysms is extremely fatal and disabling. It’s imperative for radiologists to achieve efficient screening with the help of deep learning-based models. To improve the detection sensitivity of time of flight-magnetic resonance angiography (TOF-MRA) images, this study proposed a neural network named DCAU-Net which is based on fuzzy labels, 3D U-Net variant, and dual-branch channel attention (DCA), and able to adaptively adjust the response of channel features to improve feature extraction capability. First, TOF-MRA images from 260 subjects were preprocessed, and the data were split into the training set (N=174), validation set (N=43) and testing set (N=43). Then the preprocessed data were used for training and validating DCAU-Net. The results show that DCAU-Net scores 90.69% of sensitivity, 0.83 per case of false positive count and 0.52 of positive predicted value in the testing set, providing a promising tool for detecting cerebral aneurysms.

Key words: cerebral aneurysm, automatic detection, deep learning, fuzzy label, dual branch channel attention mechanism

中图分类号:

TP394.1

陈萌, 耿辰, 李郁欣, 耿道颖, 鲍奕仿, 戴亚康. 基于模糊标签和深度学习的TOF-MRA影像脑动脉瘤自动检测[J]. 波谱学杂志, 2022, 39(3): 267-277.

Meng CHEN, Chen GENG, Yu-xin LI, Dao-ying GENG, Yi-fang BAO, Ya-kang DAI. Automatic Detection for Cerebral Aneurysms in TOF-MRA Images Based on Fuzzy Label and Deep Learning[J]. Chinese Journal of Magnetic Resonance, 2022, 39(3): 267-277.

图/表 11

表1

表2

图1

图2

图3

图4

图5

图6

图7

图8

表3

参考文献 22

1	邹应诚. 基于卷积神经网络的颅内动脉瘤检测方法研究[D]. 武汉: 华中科技大学, 2019.
2	SUAREZ J I, TARR R W, SELMAN W RAneurysmal subarachnoid hemorrhage[J].New Engl J Med,2006,354(4):387-396. doi: 10.1056/NEJMra052732
3	中国蛛网膜下腔出血诊治指南2019[J]. 中华神经科杂志, 2019, 52(12): 1006-1021.
	Chinese guidelines for diagnosis and treatment of subarachnoid hemorrhage 2019[J]. Chinese Journal of Neurology, 2019, 52(12): 1006-1021.
4	沈瑞乐. 脑动脉瘤形成、破裂与脑血管形态学类型相关性研究[D]. 郑州: 郑州大学, 2016.
5	JOSEPH J J, DONNER T WLong-term insulin glargine therapy in type 2 diabetes mellitus: a focus on cardiovascular outcomes[J].Vasc Health Risk Manag,2015,11,107-116.
6	SONOBE M, YAMAZAKI T, YONEKURA M, et alSmall unruptured intracranial aneurysm verification study: SUAVe study, Japan[J].Stroke,2010,41(9):1969-1977. doi: 10.1161/STROKEAHA.110.585059
7	娄云重, 刘颖, 江华, 等基于MRI和深度学习的桥小脑角区脑膜瘤与听神经瘤分类算法研究[J].波谱学杂志,2020,37(3):300-310.
	LOU Y Z, LIU Y, JIANG H, et alA deep learning algorithm for classifying meningioma and auditory neuroma in the cerebellopontine angle from magnetic resonance images[J].Chinese J Magn Reson,2020,37(3):300-310.
8	SAILER A M, WAGEMANS B A, NELEMANS P J, et alDiagnosing intracranial aneurysms with MR angiography: systematic review and meta-analysis[J].Stroke,2014,45(1):119-126. doi: 10.1161/STROKEAHA.113.003133
9	DAVIS W I, WARNOCK S H, HANSBERGER H RIntracranial MRA: single volume vs. multiple thin slab 3D time-of-flight acquisition[J].J Comput Assist Tomogr,1993,17(1):15-21. doi: 10.1097/00004728-199301000-00002
10	魏志宏, 闫士举, 韩宝三, 等基于多输出的3D卷积神经网络诊断阿尔兹海默病[J].波谱学杂志,2021,38(1):92-100.
	WEI Z H, YAN S J, HAN B S, et alDiagnosis of Alzheimer’s disease based on multi-output three-dimensional convolutional neural network[J].Chinese J Magn Reson,2021,38(1):92-100.
11	NAKAO T, HANAOKA S, NOMURA Y, et al Deep neural network-based computer-assisted detection of cerebral aneurysms in MR angiography[J]. J Magn Reson Imaging, 2018, 47 (4): 948- 953. doi: 10.1002/jmri.25842
12	SAKAI M, MURAYAMA S, GIBO M, et alCan maximum intensity projection images with multidetector-row computed tomography help to differentiate between the micronodular distribution of focal and diffuse infiltrative lung diseases?[J].J Comput Assist Tomogr,2005,29(5):588-591. doi: 10.1097/01.rct.0000175710.98923.a2
13	UEDA D, YAMAMOTO A, NISHIMORI M, et alDeep learning for MR angiography: Automated detection of cerebral aneurysms[J].Radiology,2019,290(1):187-194. doi: 10.1148/radiol.2018180901
14	HE K M, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition: IEEE, 2016: 770-778.
15	SICHTERMANN T, FARON A, SIJBEN R, et alDeep learning-based detection of intracranial aneurysms in 3D TOF-MRA[J].Am J Neuroradiol,2019,40(1):25-32. doi: 10.3174/ajnr.A5911
16	KAMNITSAS K, LEDIG C, NEWCOMBE VFJ, et al Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation[J]. Med Image Anal, 2017, 36, 61- 78. doi: 10.1016/j.media.2016.10.004
17	GENG C, XIA W, HUANG L, et al Automated computer-assisted detection system for cerebral aneurysms in time-of-flight magnetic resonance angiography using fully convolutional network[J]. Biomed Eng Online, 2020, 19, 38. doi: 10.1186/s12938-020-00770-7
18	ÇIçEK Ö, ABDULKADIR A, LIENKAMP S S, et al. 3D U-Net: Learning dense volumetric segmentation from sparse annotation[C]// Medical Image Computing and Computer-Assisted Intervention, Berlin, Heidelberg. USA: MICCAI, 2016: 424-432.
19	GOYAL H, SANDEEP D, VENU R, et alNormalization of data in data mining[J].International Journal of Software and Web Sciences,2014,10,32-33.
20	DIEDERIK P K, JIMMY B. Adam: A method for stochastic optimization. arXiv preprint arXiv: 1412.6980, 2014.
21	MACYSZYN L, AKBARI H, PISAPIA J M, et alImaging patterns predict patient survival and molecular subtype in glioblastoma via machine learning techniques[J].Neuro-oncolog,2015,18(3):417-425.
22	EGAN J P, GREENBERG G Z, SCHULMAN A IOperating characteristics, signal detectability, and the method of free response[J].J Acoust Soc Am,1961,33,993-1007. doi: 10.1121/1.1908935

特征	训练集	验证集	测试集
患者数量	174	43	43
年龄/岁	61±12	62±10	58±14
男性数量/女性数量	73/101	16/27	14/29
动脉瘤最大径/mm	5.48±3.29	5.42±3.25	5.48±3.82
动脉瘤总数量	182	46	46
颈内动脉区动脉瘤数量	100	25	32
大脑中动脉区动脉瘤数量	26	10	4
大脑前动脉区动脉瘤数量	30	6	7
大脑后动脉区动脉瘤数量	17	1	0
基底动脉区动脉瘤数量	5	3	3
椎动脉区动脉瘤数量	4	1	0

仪器	场强	重复时间/回波时间	视野百分比相位	采集矩阵	翻转角	层厚
Siemens	3 T	21 ms/3.69 ms	58%~90%	(256~384) × (197~331)	18?~25?	0.5~1 mm
GE	1.5 T	33 ms/6.3 ms	75%~100%	(288~384) × 192/195	20?	1.2~1.6 mm
	3 T	25 ms/3.4 ms	70%~94%	(320~384) × 192	15?	1~2.4 mm

方法	Sensitivity/%	FP count/(个/例)	PPV
ResNet-18^[14]	51.16	0.57	0.44
DeepMedic^[15]	86.04	3.40	0.21
3D U-Net^[17]	81.39	1.09	0.39
DCAU-Net（本文方法）	90.69	0.83	0.52

[1]	周勤, 王远军. 基于变分推断的磁共振图像群组配准[J]. 波谱学杂志, 2022, 39(3): 291-302.
[2]	常晓,蔡昕,杨光,聂生东. 生成对抗网络在医学图像转换领域的应用[J]. 波谱学杂志, 2022, 39(3): 366-380.
[3]	王颖珊, 邓奥琦, 毛瑾玲, 朱中旗, 石洁, 杨光, 马伟伟, 路青, 汪红志. 基于3D VNetTrans的膝关节滑膜磁共振图像自动分割[J]. 波谱学杂志, 2022, 39(3): 303-315.
[4]	王振宇, 王颖珊, 毛瑾玲, 马伟伟, 路青, 石洁, 汪红志. 基于Dense-UNet++的关节滑膜磁共振图像分割[J]. 波谱学杂志, 2022, 39(2): 208-219.
[5]	霍璐,胡晓欣,肖勤,顾雅佳,褚旭,姜娈. 基于nnU-Net的乳腺DCE-MR图像中乳房和腺体自动分割[J]. 波谱学杂志, 2021, 38(3): 367-380.
[6]	刘鹏, 钟玉敏, 王丽嘉. 基于密集多尺度U-net网络的电影心脏磁共振图像右心室自动分割[J]. 波谱学杂志, 2020, 37(4): 456-468.
[7]	赵尚义, 王远军. 基于磁共振图像和改进的UNet++模型区分阿尔茨海默症患者和健康人群[J]. 波谱学杂志, 2020, 37(3): 321-331.
[8]	宫进昌, 王宇, 王远军. 结合小波融合和深度学习的脑胶质瘤自动分割[J]. 波谱学杂志, 2020, 37(2): 131-143.
[9]	程慧涛, 王珊珊, 柯子文, 贾森, 程静, 丘志浪, 郑海荣, 梁栋. 基于深度递归级联卷积神经网络的并行磁共振成像方法[J]. 波谱学杂志, 2019, 36(4): 437-445.
[10]	苏新宇, 王丽嘉, 聂生东, 胡立伟, 钟玉敏. 基于心脏磁共振短轴电影图像的右心室分割新进展[J]. 波谱学杂志, 2019, 36(3): 377-391.

基于模糊标签和深度学习的TOF-MRA影像脑动脉瘤自动检测

Automatic Detection for Cerebral Aneurysms in TOF-MRA Images Based on Fuzzy Label and Deep Learning

RichHTML

PDF

评审附件

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 22

相关文章 10

编辑推荐

Metrics

本文评价