《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (9): 2909-2916.DOI: 10.11772/j.issn.1001-9081.2021071206
刘汉卿1, 康晓东1(), 张福青2, 赵秀圆2, 杨靖怡1, 王笑天1, 李梦凡3
收稿日期:
2021-07-12
修回日期:
2021-09-15
接受日期:
2021-09-22
发布日期:
2022-09-19
出版日期:
2022-09-10
通讯作者:
康晓东
作者简介:
刘汉卿(1997—),男,湖南衡阳人,硕士研究生,主要研究方向:医学图像处理;基金资助:
Hanqing LIU1, Xiaodong KANG1(), Fuqing ZHANG2, Xiuyuan ZHAO2, Jingyi YANG1, Xiaotian WANG1, Mengfan LI3
Received:
2021-07-12
Revised:
2021-09-15
Accepted:
2021-09-22
Online:
2022-09-19
Published:
2022-09-10
Contact:
Xiaodong KANG
About author:
LIU Hanqing, born in 1997, M. S. candidate. His research interests include medical image processing.Supported by:
摘要:
针对断层面上血管的多形性和检测过程中出现的采样不均衡的问题,提出一种改进的Libra区域卷积神经网络(R-CNN)的脑动脉狭窄影像学检测算法,用于检测计算机断层扫描血管造影(CTA)图像的颈内动脉和椎动脉狭窄。首先,在目标检测网络Libra R-CNN中以ResNet50为骨干网络,并分别在骨干网络的3、4、5阶段引入可变卷积网络(DCN),通过学习偏移量提取血管在不同断层面的形态特征;然后,将从骨干网络中提取的特征图输入至引入非局部神经网络(Non-local NN)的平衡特征金字塔(BFP)中进行更深度的特征融合;最后,将融合后的特征图输入至级联检测器,并通过提高交并比(IoU)阈值优化最终检测结果。实验结果表明,改进的Libra R-CNN检测算法相比Libra R-CNN算法,在脑动脉CTA数据集中平均准确率(AP)、AP50、AP75和APS分别提升了4.3、1.3、6.9和4.0个百分点;在公开的结肠息肉CT数据集中,AP、AP50、AP75和APS分别提升了6.6、3.6、13.0和6.4个百分点。通过在Libra R-CNN的骨干网络中加入DCN、Non-local NN和级联检测器,进一步融合特征从而学习脑动脉血管结构的语义信息,使得狭窄区域检测结果更精确,且改进算法在不同的检测任务中具有泛化能力。
中图分类号:
刘汉卿, 康晓东, 张福青, 赵秀圆, 杨靖怡, 王笑天, 李梦凡. 改进的Libra区域卷积神经网络的脑动脉狭窄影像学检测算法[J]. 计算机应用, 2022, 42(9): 2909-2916.
Hanqing LIU, Xiaodong KANG, Fuqing ZHANG, Xiuyuan ZHAO, Jingyi YANG, Xiaotian WANG, Mengfan LI. Image detection algorithm of cerebral arterial stenosis by improved Libra region-convolutional neural network[J]. Journal of Computer Applications, 2022, 42(9): 2909-2916.
算法 | Backbone | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|---|
Faster R-CNN[ | ResNet50 | 37.7 | 47.3 | 96.2 | 31.3 | 47.3 |
YOLOv3[ | ResNet50 | 56.2 | 43.9 | 95.9 | 24.6 | 43.9 |
Libra R-CNN[ | ResNet50 | 33.7 | 47.6 | 96.0 | 35.8 | 47.6 |
Cascade R-CNN[ | ResNet50 | 26.1 | 49.8 | 94.8 | 41.8 | 49.8 |
本文算法 | ResNet50 | 22.9 | 51.9 | 97.3 | 42.7 | 51.6 |
表1 实验1的对比结果
Tab.1 Comparison results of experiment 1
算法 | Backbone | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|---|
Faster R-CNN[ | ResNet50 | 37.7 | 47.3 | 96.2 | 31.3 | 47.3 |
YOLOv3[ | ResNet50 | 56.2 | 43.9 | 95.9 | 24.6 | 43.9 |
Libra R-CNN[ | ResNet50 | 33.7 | 47.6 | 96.0 | 35.8 | 47.6 |
Cascade R-CNN[ | ResNet50 | 26.1 | 49.8 | 94.8 | 41.8 | 49.8 |
本文算法 | ResNet50 | 22.9 | 51.9 | 97.3 | 42.7 | 51.6 |
算法 | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|
Libra R-CNN | 33.7 | 47.6 | 96.0 | 35.8 | 47.6 |
Libra R-CNN+DCN | 32.0 | 47.8 | 95.2 | 36.8 | 47.9 |
Libra R-CNN+DCN+Cascade | 22.7 | 50.5 | 95.9 | 42.8 | 50.8 |
Libra R-CNN+DCN+Cascade+Non-Local NN | 22.9 | 51.9 | 97.3 | 42.7 | 51.6 |
表2 实验1的消融结果
Tab.2 Ablation results of experiment 1
算法 | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|
Libra R-CNN | 33.7 | 47.6 | 96.0 | 35.8 | 47.6 |
Libra R-CNN+DCN | 32.0 | 47.8 | 95.2 | 36.8 | 47.9 |
Libra R-CNN+DCN+Cascade | 22.7 | 50.5 | 95.9 | 42.8 | 50.8 |
Libra R-CNN+DCN+Cascade+Non-Local NN | 22.9 | 51.9 | 97.3 | 42.7 | 51.6 |
算法 | Backbone | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|---|
Faster R-CNN[ | ResNet50 | 36.7 | 52.0 | 94.1 | 54.0 | 53.8 |
YOLOv3[ | ResNet50 | 57.4 | 47.2 | 92.2 | 47.0 | 37.8 |
Libra R-CNN[ | ResNet50 | 33.2 | 53.2 | 95.7 | 53.9 | 53.4 |
Cascade R-CNN[ | ResNet50 | 24.8 | 56.1 | 91.0 | 63.4 | 56.1 |
本文算法 | ResNet50 | 22.3 | 59.8 | 99.3 | 66.9 | 59.8 |
表3 实验2的对比结果
Tab.3 Comparison results of experiment 2
算法 | Backbone | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|---|
Faster R-CNN[ | ResNet50 | 36.7 | 52.0 | 94.1 | 54.0 | 53.8 |
YOLOv3[ | ResNet50 | 57.4 | 47.2 | 92.2 | 47.0 | 37.8 |
Libra R-CNN[ | ResNet50 | 33.2 | 53.2 | 95.7 | 53.9 | 53.4 |
Cascade R-CNN[ | ResNet50 | 24.8 | 56.1 | 91.0 | 63.4 | 56.1 |
本文算法 | ResNet50 | 22.3 | 59.8 | 99.3 | 66.9 | 59.8 |
算法 | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|
Libra R-CNN | 33.2 | 53.2 | 95.7 | 53.9 | 53.4 |
Libra R-CNN+DCN | 32.1 | 53.9 | 91.4 | 55.0 | 53.3 |
Libra R-CNN+DCN+Cascade | 23.0 | 55.6 | 92.5 | 58.2 | 55.6 |
Libra R-CNN+DCN+ Cascade+Non-Local | 22.3 | 59.8 | 99.3 | 66.9 | 59.8 |
表4 实验2的消融结果
Tab.4 Ablation results of experiment 2
算法 | FPS | AP/% | AP50/% | AP75/% | APS/% |
---|---|---|---|---|---|
Libra R-CNN | 33.2 | 53.2 | 95.7 | 53.9 | 53.4 |
Libra R-CNN+DCN | 32.1 | 53.9 | 91.4 | 55.0 | 53.3 |
Libra R-CNN+DCN+Cascade | 23.0 | 55.6 | 92.5 | 58.2 | 55.6 |
Libra R-CNN+DCN+ Cascade+Non-Local | 22.3 | 59.8 | 99.3 | 66.9 | 59.8 |
1 | HSIEH Y Z, LUO Y C, PAN C, et al. Cerebral small vessel disease biomarkers detection on MRI-sensor-based image and deep learning[J]. Sensors, 2019, 19(11): No.2573. 10.3390/s19112573 |
2 | BASH S, VILLABLANCA J P, JAHAN R, et al. Intracranial vascular stenosis and occlusive disease: evaluation with CT angiography, MR angiography, and digital subtraction angiography[J]. American Journal of Neuroradiology, 2005, 26(5): 1012-1021. |
3 | WADDLE S L, JUTTUKONDA M R, LANTS S K, et al. Classifying intracranial stenosis disease severity from functional MRI data using machine learning[J]. Journal of Cerebral Blood Flow and Metabolism, 2020, 40(4): 705-719. 10.1177/0271678x19848098 |
4 | HSU K C, LIN C H, JOHNSON K R, et al. Autodetect extracranial and intracranial artery stenosis by machine learning using ultrasound[J]. Computers in Biology and Medicine, 2020, 116: No.103569. 10.1016/j.compbiomed.2019.103569 |
5 | ARAKI T, JAIN P K, SURI H S, et al. Stroke risk stratification and its validation using ultrasonic echolucent carotid wall plaque morphology: a machine learning paradigm[J]. Computers in Biology and Medicine, 2017, 80:77-96. 10.1016/j.compbiomed.2016.11.011 |
6 | SABA L, JAIN P K, SURI H S, et al. Plaque tissue morphology-based stroke risk stratification using carotid ultrasound: a polling-based PCA learning paradigm[J]. Journal of Medical Systems, 2017, 41(6): No.98. 10.1007/s10916-017-0745-0 |
7 | HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. 10.1109/tpami.2015.2389824 |
8 | GIRSHICK R. Fast R-CNN[C]// Proceedings of the 2015 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2015: 1440-1448. 10.1109/iccv.2015.169 |
9 | REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[C]// Proceedings of the 28th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2015:91-99. |
10 | LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 936-944. 10.1109/cvpr.2017.106 |
11 | JOO B, AHN S S, YOON P H, et al. A deep learning algorithm may automate intracranial aneurysm detection on MR angiography with high diagnostic performance[J]. European Radiology, 2020, 30(11):5785-5793. 10.1007/s00330-020-06966-8 |
12 | SMISTAD E, LØVSTAKKEN L. Vessel detection in ultrasound images using deep convolutional neural networks[C]// Proceedings of the 2016 International Workshop on Deep Learning in Medical Image Analysis/2016 International Workshop on Large-Scale Annotation of Biomedical Data and Expert Label Synthesis, LNCS 10008. Cham: Springer, 2016: 30-38. |
13 | STIB M T, VASQUEZ J, DONG M P, et al. Detecting large vessel occlusion at multiphase CT angiography by using a deep convolutional neural network[J]. Radiology, 2020, 297(3): 640-649. 10.1148/radiol.2020200334 |
14 | DE MAN Q, HANEDA E, CLAUS B, et al. A two-dimensional feasibility study of deep learning-based feature detection and characterization directly from CT sinograms[J]. Medical Physics, 2019, 46(12): e790-e800. 10.1002/mp.13640 |
15 | 胡涛,苏佳斌,杨恒,等. 基于注意力机制的CNN和GRU烟雾病检测方法研究[J]. 航天医学与医学工程, 2021, 34(1):68-74. 10.16289/j.cnki.1002-0837.2021.01.011 |
HU T, SU J B, YANG H, et al. Research on detection of moyamoya disease by CNN and GRU based on attention mechanism[J]. Space Medicine and Medical Engineering, 2021, 34(1):68-74. 10.16289/j.cnki.1002-0837.2021.01.011 | |
16 | DAI X L, HUANG L X, QIAN Y, et al. Deep learning for automated cerebral aneurysm detection on computed tomography images[J]. International Journal of Computer Assisted Radiology and Surgery, 2020, 15(4): 715-723. 10.1007/s11548-020-02121-2 |
17 | YANG J H, XIE M F, HU C P, et al. Deep learning for detecting cerebral aneurysms with CT angiography[J]. Radiology, 2021, 298(1): 155-163. 10.1148/radiol.2020192154 |
18 | SHINOHARA Y, TAKAHASHI N, LEE Y, et al. Development of a deep learning model to identify hyperdense MCA sign in patients with acute ischemic stroke[J]. Japanese Journal of Radiology, 2020, 38(2): 112-117. 10.1007/s11604-019-00894-4 |
19 | 卫渊,朱光宇,颜格,等. 基于卷积神经网络的颅内动脉瘤CTA影像判别[J]. 医用生物力学, 2019, 34(S1):147. |
WEI Y, ZHU G Y, YAN G, et al. CTA image discrimination of intracranial aneurysm based on convolutional neural network[J]. Journal of Medical Biomechanics, 2019, 34(S1): 147. | |
20 | 秦志光,陈浩,丁熠,等. 基于多模态卷积神经网络的脑血管提取方法研究[J]. 电子科技大学学报, 2016, 45(4):573-581. 10.3969/j.issn.1001-0548.2016.04.010 |
QIN Z G, CHEN H, DING Y, et al. Research on brain vessel extraction via multi-modal convolutional neural networks[J]. Journal of University of Electronic Science and Technology of China, 2016, 45(4): 573-581. 10.3969/j.issn.1001-0548.2016.04.010 | |
21 | PANG J M, CHEN K, SHI J P, et al. Libra R-CNN: towards balanced learning for object detection[C]// Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2019: 821-830. 10.1109/cvpr.2019.00091 |
22 | WANG X L, GIRSHICK R, GUPTA A, et al. Non-local neural networks[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 7794-7803. 10.1109/cvpr.2018.00813 |
23 | SHOKRI M, HARATI A, TABA K. Salient object detection in video using deep non-local neural networks[J]. Journal of Visual Communication and Image Representation, 2020, 68: No.102769. 10.1016/j.jvcir.2020.102769 |
24 | CAI Z W, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 6154-6162. 10.1109/cvpr.2018.00644 |
25 | REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. (2018-04-08) [2021-08-15].. |
[1] | 况世雄 姚俊波 陆佳炜 王琪冰 肖刚. 基于动态图卷积网络的电梯乘客异常行为数据增强方法[J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[2] | 康斌 陈斌 王俊杰 李昱林 赵军智 咸伟志. 基于多粒度共享语义中心关联的文本到人物检索方法[J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[3] | 王子怡 李卫军 刘雪洋 丁建平 刘世侠 苏易礌. 基于Swin Transformer与多尺度特征融合的图像描述方法#br# [J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[4] | 付可意, 王高才, 邬满. 基于改进区域提议网络和特征聚合小样本目标检测方法[J]. 《计算机应用》唯一官方网站, 2024, 44(12): 3790-3797. |
[5] | 庞玉东, 李志星, 刘伟杰, 李天昊, 王宁宁. 基于改进实时检测Transformer的塔机上俯视场景小目标检测模型[J]. 《计算机应用》唯一官方网站, 2024, 44(12): 3922-3929. |
[6] | 赵欣, 李鑫杰, 徐健, 刘步云, 毕祥. 基于卷积神经网络与Transformer并行的医学图像配准模型[J]. 《计算机应用》唯一官方网站, 2024, 44(12): 3915-3921. |
[7] | 颜承志 陈颖 钟凯 高寒. 基于多尺度网络与轴向注意力的3D目标检测算法[J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[8] | 王静 刘嘉星 宋婉莹 薛嘉兴 丁温欣. 基于空间变换和特征分布校准的小样本皮肤图像分类模型[J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[9] | 廖炎华 鄢元霞 潘文林. 基于YOLOv9的交通路口图像的多目标检测算法[J]. 《计算机应用》唯一官方网站, 0, (): 0-0. |
[10] | 谢斌红 剌颖坤 张英俊 张睿. 自步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, 0, (): 0-0. |
[11] | 邹耀斌, 张彬. 四向加权香农熵最大化导向的自动阈值分割方法[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3565-3573. |
[12] | 刘涛, 鞠事宏, 高一萌. 基于改进YOLOv8n的无人机视角下小目标检测算法[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3603-3609. |
[13] | 刘雨生, 肖学中. 基于扩散模型微调的高保真图像编辑[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3574-3580. |
[14] | 胡立华, 李小平, 胡建华, 张素兰. 基于四叉树先验辅助的多视图立体方法[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3556-3564. |
[15] | 顾聪, 段其强, 任思雨. 基于上下文感知网络的息肉分割算法[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3617-3622. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||