基于3D ResNet50改进模型的TOF-MRA脑动脉瘤分类方法

doi:10.11938/cjmr20243119

摘要/Abstract

摘要：

脑动脉瘤的不规则形态，尤其是子瘤的存在，是动脉瘤破裂风险的关键因素．临床上对子瘤的评估主要是通过时间飞跃法磁共振血管造影（Time of Flight-Magnetic Resonance Angiography，TOF-MRA）进行图像重建及基于医生视觉和经验的判断，这限制了诊断的效率和准确性．本文提出了一种基于3D ResNet50改进的并行多尺度注意力融合网络（Parallel Multiscale Attention Fusion Networks，PMAF-Net）的子瘤自动分类方法，PMAF-Net采用多尺度卷积并加权融合通道和空间注意力权重以提高特征提取能力．实验所用TOF-MRA数据291例，其中训练集128例，验证集32例，测试集131例．与其他分类网络比较，PMAF-Net在测试集上表现最好，准确率为83.97%，召回率为84.48%，精确率为80.33%，F1分数为0.823 5，受试者工作特征曲线（ROC）也显示出模型最佳的分类性能（AUC为0.900 8）．实验结果表明该网络能更准确地识别出子瘤型动脉瘤，有望对动脉瘤破裂风险评估和量化提供支持．

关键词: 时间飞跃法磁共振血管造影, 脑动脉瘤子瘤, 多尺度卷积注意力, 自动分类, 深度学习

Abstract:

The irregular morphology of cerebral aneurysms, especially the presence of a daughter sac, is a crucial risk factor for aneurysm rupture. Clinical assessment of daughter sac relies mainly on image reconstruction by time of flight-magnetic resonance angiography (TOF-MRA) and judgment based on physicians' vision and experience, which limits the efficiency and accuracy of diagnosis. In this paper, we propose an improved parallel multiscale fusion attention network (PMAF-Net) based on 3D ResNet50 for classification. PMAF-Net uses multi-scale convolution and weighted fusion channel and spatial attention weights to enhance the feature extraction capability. The experiment used 291 cases of TOF-MRA data, including 128 cases in the training set, 32 cases in the validation set, and 131 cases in the test set. Compared with other classification networks, PMAF-Net performs best on the test set, with the accuracy of 83.97%, recall of 84.48%, precision of 80.33%, and F1-score of 0.823 5, and the receiver operating characteristic curve (ROC) also reflects the model's optimal classification performance (AUC of 0.900 8). The results show that the network can identify daughter sac type aneurysms more accurately, which is expected to support the assessment and quantification of the risk of aneurysm rupture.

Key words: TOF-MRA, daughter sac of cerebral aneurysms, multiscale convolution attention, automatic classification, deep learning

中图分类号:

O482.53

薛培阳, 耿辰, 李郁欣, 鲍奕仿, 鲁宇澄, 戴亚康. 基于3D ResNet50改进模型的TOF-MRA脑动脉瘤分类方法[J]. 波谱学杂志, 2025, 42(1): 56-66.

XUE Peiyang, GENG Chen, LI Yuxin, BAO Yifang, LU Yucheng, DAI Yakang. A Classification Method for Cerebral Aneurysms in TOF-MRA Based on Improved 3D ResNet50 Model[J]. Chinese Journal of Magnetic Resonance, 2025, 42(1): 56-66.

图/表 11

图1

图2

图3

表1

表2

图4

表3

图5

表4

图6

表5

参考文献 28

[1]	WU Y Y, ZHANG L J, JING W W, et al. Research progress in the application of radiomics in intracranial aneurysms[J]. Chin J Cerebrovasc Dis, 2023, 20(11): 769-776.
	吴钖莹, 张丽娟, 敬维维, 等. 影像组学在颅内动脉瘤中的应用研究进展[J]. 中国脑血管病杂志, 2023, 20(11): 769-776.
[2]	FORBES G, FOX A J, HUSTON J, et al. Interobserver variability in angiographic measurement and morphologic characterization of intracranial aneurysms: a report from the International Study of Unruptured Intracranial Aneurysms[J]. Am J Neuroradiol, 1996, 17(8): 1407-1415. pmid: 8883634
[3]	MORITA A, KIRINO T, HASHI K, et al. The Natural Course of Unruptured Cerebral Aneurysms in a Japanese Cohort[J]. N Engl J Med, 2012, 366(26): 2474-2482.
[4]	VIAK M H M, ALGRA A, BRANDENBURG R, et al. Prevalence of unruptured intracranial aneurysms, with emphasis on sex, age, comorbidity, country, and time period: a systematic review and meta-analysis[J]. Lancet Neurol, 2011, 10(7): 626-636. doi: 10.1016/S1474-4422(11)70109-0 pmid: 21641282
[5]	ISHIBASHI T, MURAYAMA Y, URASHIMA M, et al. Unruptured intracranial aneurysms incidence of rupture and risk factors[J]. Stroke, 2009, 40(1): 313-6.
[6]	GORINI S, QUIRINI M, MENCIASSI A, et al. A novel SMA-based actuator for a legged endoscopic capsule[C]// International Conference on Biomedical Robotics and Biomechatronics, 2006: 443-449.
[7]	WERMER M J H, VAN DER SCHAAF I C, ALGRA A, et al. Risk of rupture of unruptured intracranial aneurysms in relation to patient and aneurysm characteristics - An updated meta-analysis[J]. Stroke, 2007, 38(4): 1404-1410.
[8]	TSUKAHARA T, Murakami N, SAKURAI Y, et al. Treatment of unruptured cerebral aneurysms; a multi-center study at Japanese national hospitals[C]// New Trends of Surgery for Stroke and its Perioperative Management. Springer Vienna, 2005: 77-85.
[9]	BACKES D, RINKEL G J E, Greving J P, et al. ELAPSS score for prediction of risk of growth of unruptured intracranial aneurysms[J]. JAMA Neurol, 2017, 88(17): 1600-1606.
[10]	ZHANG J J, LU Y C, BAO Y F, et al. An automatic segmentation method of cerebral arterial tree in TOF-MRA based on DBCNet[J]. Chinese J Magn Reson, 2023, 40(3): 320-331.
	张嘉骏, 鲁宇澄, 鲍奕仿, 等. 基于DBCNet的TOF-MRA中脑动脉树区域自动分割方法[J]. 波谱学杂志, 2023, 40(3): 320-331. doi: 10.11938/cjmr20223046
[11]	CHEN G, BAO Y F, ZHANG J J, et al. Automated unruptured cerebral aneurysms detection in TOF MR angiography images using dual-channel SE-3D UNet: a multi-center research[J]. Eur Radiol, 2023, 33(5): 3532-3543. doi: 10.1007/s00330-022-09385-z pmid: 36725720
[12]	ASHKEZARI S F S, DETMER F J, MUT F, et al. Blebs in intracranial aneurysms: prevalence and general characteristics[J]. J Neurointerv Surg, 2021, 13(3): 226-230. doi: 10.1136/neurintsurg-2020-016274 pmid: 32680877
[13]	CHEN X, WANG X, ZhANG K, et al. Recent advances and clinical applications of deep learning in medical image analysis[J]. Med Image Anal, 2022, 79: 102444.
[14]	ZHOU S K, GREENSPAN H, DAVATZIKOS C, et al. A review of deep learning in medical imaging: Imaging traits, technology trends, case studies with progress highlights, and future promises[J]. Proc IEEE Inst Electr Electron Eng, 2021, 109(5): 820-838.
[15]	UEDA D, YAMAMOTO A, NISHIMORI M, et al. Deep learning for MR angiography: automated detection of cerebral aneurysms[J]. Radiology, 2019, 290(1): 187-194. doi: 10.1148/radiol.2018180901 pmid: 30351253
[16]	HE J Q, AN X W, REN H C, et al. Progresses of computer aided diagnosis in detection of intracranial aneurysms[J]. Chin J Med Imaging Technol, 2023, 39(2): 291-294.
	何佳倩, 安兴伟, 任贺成, 等. 计算机辅助诊断检测颅内动脉瘤进展[J]. 中国医学影像技术, 2023, 39(2): 291-294.
[17]	ZHAO Y, WANG S, REN Y, et al. CRANet: a comprehensive residual attention network for intracranial aneurysm image classification[J]. BMC Bioinform, 2022, 23(1): 322.
[18]	HU Y, XU Y, HUANG X, et al. CARNet: Automatic Cerebral Aneurysm Classification in Time-of-Flight MR Angiography by Leveraging Recurrent Neural Networks[C]// First CAAI International Conference, Springer International Publishing, 2021: 136-148.
[19]	CHEN M, GENG C, LI Y X, et al. Automatic detection for cerebral aneurysms in TOF-MRA images based on fuzzy label and deep learning[J]. Chinese J Magn Reson, 2022, 39(3): 267-277.
	陈萌, 耿辰, 李郁欣, 等. 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. doi: 10.11938/cjmr20223004
[20]	HARA K, KATAOKA H, SATOH Y. Learning spatio-temporal features with 3d residual networks for action recognition[C]// International Conference on Computer Vision Workshops. 2017: 3154-3160.
[21]	YANG J, SHI R, WEI D, et al. Medmnist v2-a large-scale lightweight benchmark for 2d and 3d biomedical image classification[J]. Sci Data, 2023, 10(1): 41.
[22]	HASSAN E, HOSSAIN M S, SABER A, et al. A quantum convolutional network and ResNet (50)-based classification architecture for the MNIST medical dataset[J]. Biomed Signal Process Control, 2024, 87: 105560.
[23]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Conference Vision and Pattern Recognition, 2016: 770-778.
[24]	WOO S, PARK J, LEE J Y, et al. CBAM: Convolutional block attention module[C]// Computer Vision-ECCV 2018, Lecture Notes in Computer Science, 2018, 11211: 3-19.
[25]	JI Z, GENG J W, ZHAI X D, et al. Expert consensus on imaging interpretation of intracranial aneurysms[J]. Chin J Cerebrovasc Dis, 2021, 18(7):492-504.
	吉喆, 耿介文, 翟晓东, 等. 颅内动脉瘤影像学判读专家共识[J]. 中国脑血管病杂志, 2021, 18(7): 492-504.
[26]	HUANG G, LIU S, VAN DER MAATEN L, et al. Condensenet: An efficient densenet using learned group convolutions[C]// Conference Vision and Pattern Recognition, 2018: 2752-2761.
[27]	SANDLER M, HOWARD A, ZHU M, et al. Mobilenetv2: Inverted residuals and linear bottlenecks[C]// Conference Vision and Pattern Recognition, 2018: 4510-4520.
[28]	TAN M, LE Q. Efficientnet: Rethinking model scaling for convolutional neural networks[C]// International Conference on Machine Learning, 2019: 6105-6114.

组别	训练集	验证集	测试集
患者数量	128	32	131
男性/女性	56/72	12/20	71/60
年龄（岁）	56±12	58±9	62±14
正常动脉瘤/子瘤型动脉瘤	78/50	19/13	76/55
动脉瘤大小（mm）	6.75±5.23	6.04±4.21	6.92±6.27
小型动脉瘤（<5mm）/子瘤	46/10	11/4	47/9
中型动脉瘤（5≤最大径<15mm）/子瘤	74/37	19/8	81/46
大型动脉瘤（15≤最大径<25mm）/子瘤	8/3	2/1	3/0

采集设备	重复时间	回波时间	图像分辨率	层厚	翻转角	采集矩阵	扫描时间
GE 3.0T	25 ms	5.7 ms	1024×1024×240	1.2 mm	20˚	320×256	3min22s
SIEMENS 3.0T	21 ms	3.6 ms	512×512×128	0.9 mm	18˚	256×197	2min53s
Philips 3.0T	18 ms	3.5 ms	512×512×128	1.0 mm	20˚	308×203	2min22s

类别 Category	精确率 Precision	召回率 Recall	F1分数 F1_Score
子瘤型动脉瘤（有子瘤）	80.33%	84.48%	0.8235
正常动脉瘤（无子瘤）	87.14%	83.56%	0.8531
平均	83.74%	84.02%	0.8383

亚组分析 Subgroup analysis	准确率 Accuracy	召回率 Recall	精确率 Precision	F1分数 F1_Score
总体性能	83.97%	84.48%	80.33%	0.8235
按最大径划分
小型动脉瘤（47）	80.85%	77.78%	50.00%	0.6087
中型动脉瘤（81）	85.19%	84.78%	88.64%	0.8667
大型动脉瘤（3）	100.00%	100.00%	100.00%	1.0000

模型 Models	准确率 Accuracy	召回率 Recall	精确率 Precision	F1分数 F1_Socre
ResNet18^[23]	74.81%	60.34%	77.78%	0.6796
ResNet50^[20]	72.52%	74.14%	67.19%	0.7049
DenseNet^[26]	67.94%	68.97%	62.50%	0.6557
MobileNet^[27]	77.86%	81.03%	72.31%	0.7642
EfficientNet^[28]	72.52%	65.52%	70.37%	0.6786
PMAF-Net（本文）	83.97%	84.48%	80.33%	0.8235