广西师范大学学报（自然科学版） ›› 2025, Vol. 43 ›› Issue (6): 92-106.doi: 10.16088/j.issn.1001-6600.2024123003

• 智能信息处理 • 上一篇    下一篇

融合动态通道剪枝的轻量级CT图像肺结节检测网络设计

易见兵1,2*, 胡雅怡1,2, 曹锋1,2, 李俊1,2, 彭鑫1,2, 陈鑫1,2   

  1. 1.江西理工大学 信息工程学院, 江西 赣州 341000;
    2.多维智能感知与控制江西省重点实验室(江西理工大学), 江西 赣州 341000
  • 收稿日期:2024-12-30 修回日期:2025-05-15 发布日期:2025-11-19
  • 通讯作者: 易见兵(1980—), 男, 江西宜春人, 江西理工大学副教授, 博士。E-mail: yijianbing8@jxust.edu.cn
  • 基金资助:
    国家自然科学基金(62066018, 62366017); 江西省自然科学基金(20181BAB202004); 江西省研究生创新专项资金(YC2023-S662)

Design of Lightweight Pulmonary Nodules Detection Network on CT Images with Dynamic Channel Pruning

YI Jianbing1,2*, HU Yayi1,2, CAO Feng1,2, LI Jun1,2, PENG Xin1,2, CHEN Xin1,2   

  1. 1. College of Information Engineering, Jiangxi University of Science and Technology, Ganzhou Jiangxi 341000, China;
    2. Jiangxi Provincial Key Laboratory of Multidimensional Intelligent Perception and Control (Jiangxi University of Science and Technology), Ganzhou Jiangxi 341000, China
  • Received:2024-12-30 Revised:2025-05-15 Published:2025-11-19

PDF (PC)

1

摘要： 肺癌是全球致死率最高的肿瘤疾病,而肺结节是早期肺癌的主要表现,现有算法在检测小目标肺结节时存在漏检、误检且模型复杂度高等问题。针对以上问题,本文提出一种基于通道掩码和动态通道剪枝的肺结节检测算法。首先,利用跨层连接将浅层特征与深层特征进行融合,并且精简路径聚合网络,减少模型参数量的同时获得更丰富的特征。其次,在残差连接中引入SE通道注意力机制,通过自适应调整每个通道的权重,聚焦肺结节中的关键信息,以提高算法对肺结节的检测能力。最后,利用通道掩码对网络进行动态通道剪枝,使网络能够完整保留模块中的跳跃连接,以增强模型的特征表达能力。在LUNA16数据集上,本文算法比YOLOv8n模型权重小0.3 MiB,且召回率和mAP@0.5分别提升2.0和1.7个百分点。在Lung-PET-CT-Dx数据集上,本文算法比YOLOv8n模型权重小0.9 MiB,且召回率和mAP@0.5分别提升0.8和0.4个百分点。实验结果表明,本文模型具有较高的肺结节检测精度且参数量较少。

关键词: 肺结节检测, 通道掩码, 动态通道剪枝, 跨层特征融合, 通道注意力

Abstract: Lung cancer is the most deadly cancer in the world, and pulmonary nodules are the main manifestation of early-stage lung cancer. However, existing algorithms have problems with missed detections, false positives and high model complexity when detecting small lung tumors. To solve the above problems, a pulmonary nodules detection algorithm based on channel mask and dynamic channel pruning is proposed. Firstly, the algorithm uses cross-layer connection to fuse shallow features with deep features, and simplifies the path aggregation network, thereby obtaining more abundant features while reducing the number of model parameters. Secondly, the algorithm introduces the SE channel attention mechanism in the residual connection, and adaptively adjusting the weight of each channel to focus on the key information in the lesion in order to improve the detection ability of the algorithm for pulmonary lesions. Finally, the algorithm uses a channel mask for dynamic channel pruning, which completely retains the skip connections in the module, thereby enhancing the feature representation ability of the model. The proposed algorithm is compared with the YOLOv8n algorithm in the performance test on the LUNA16 dataset. The model weight of this algorithm is 0.3 MiB smaller than that of the YOLOv8n algorithm, and the recall rate and mAP@0.5 increase by 2.0 percentage points and 1.7 percentage points, respectively. The proposed algorithm is compared with the YOLOv8n algorithm in the performance test on the Lung-PET-CT-Dx dataset. The model weight of this algorithm is 0.9 MiB smaller than that of YOLOv8n algorithm, and the recall rate and mAP@0.5 increase by 0.8 percentage points and 0.4 percentage points, respectively. The experimental results show that the proposed model has the characteristics of higher accuracy and fewer parameters in lung cancer detection.

Key words: lung cancer detection, channel mask, dynamic channel pruning, cross-layer feature fusion, channel attention

中图分类号:  TP391.41

[1] World Health Organization. Global cancer burden growing, amidst mounting need for services[R/OL]. (2024-02-01)[2024-12-30]. https://www.who.int/news/item/01-02-2024-global-cancer-burden-growing-amidst-mounting-need-for-services.
[2] 孙晨皓, 赖惠成, 李亚东, 等. 改进YOLOX在行人目标检测中的应用研究[J]. 东北师大学报(自然科学版), 2023. 55(2):76-84. DOI: 10.16163/j.cnki.dslkxb202112030001.
[3] 催海彬, 蒲东兵, 陆云凤, 等. 基于CA-YOLO的安全帽检测[J]. 东北师大学报(自然科学版), 2023. 55(3): 94-100. DOI: 10.16163/j.cnki.dslkxb202211030001.
[4] ZHANG Y F, ZHU Z Y, HOU J H, et al. Spatial-temporal graph enhanced DETR towards multi-frame 3D object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12): 10614-10628. DOI: 10.1109/TPAMI.2024.3443335.
[5] LI M Y, JIA T, WANG H,et al. AO-DETR: anti-overlapping DETR for X-ray prohibited items detection[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(7): 12076-12090. DOI: 10.1109/TNNLS.2024.3487833.
[6] LI Z L, XU X G, LIM S, et al.UniMODE: unified monocular 3D object detection[C] //2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2024: 16561-16570. DOI: 10.1109/CVPR52733.2024.01567.
[7] LI X, LV C Q, WANG W H, et al. Generalized focal loss: towards efficient representation learning for dense object detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(3): 3139-3153. DOI: 10.1109/TPAMI.2022.3180392.
[8] LIU K, FU Z H, JIN S, et al. ESOD: efficient small object detection on high-resolution images[J]. IEEE Transactions on Image Processing, 2025, 34: 183-195. DOI: 10.1109/TIP.2024.3501853.
[9] SOGANCIOGLU E, VAN GINNEKEN B, BEHRENDT F, et al. Nodule detection and generation on chest X-rays: NODE21 challenge[J]. IEEE Transactions on Medical Imaging, 2024, 43(8): 2839-2853. DOI: 10.1109/TMI.2024.3382042.
[10] 李阳, 高轼奇. 基于数据增强及注意力机制的肺结节检测系统[J]. 北京邮电大学学报, 2022, 45(4): 25-30. DOI: 10.13190/j.jbupt.2021-189.
[11] XU J, REN H J, CAI S Z, et al. An improved faster R-CNN algorithm for assisted detection of lung nodules[J]. Computers in Biology and Medicine, 2023, 153: 106470. DOI: 10.1016/j.compbiomed.2022.106470.
[12] 杨嘉楠, 王忠昊, 王昊霖, 等. 改进BYOL的非小细胞肺癌表皮生长因子受体基因突变预测[J]. 光学精密工程, 2022, 30(9): 1080-1090. DOI: 10.37188/OPE.20223009.1080.
[13] GUO Z T, ZHAO LL, YUAN J L, et al. MSANet: multiscale aggregation network integrating spatial and channel information for lung nodule detection[J]. IEEE Journal of Biomedical and Health Informatics, 2022, 26(6): 2547-2558. DOI: 10.1109/JBHI.2021.3131671.
[14] 周涛,叶鑫宇,刘凤珍,等. 利用跨模态轻量级YOLOv5模型的PET/CT肺部肿瘤检测[J]. 电子与信息学报, 2024, 46(2): 624-632. DOI: 10.11999/JEIT230052.
[15] GUAN Y S, LIU N, ZHAO P Y, et al.DAIS: automatic channel pruning via differentiable annealing indicator search[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(12): 9847-9858. DOI: 10.1109/TNNLS.2022.3161284.
[16] CHERAGHIAN A, RAHMAN S, FANG P F, et al. Semantic-aware knowledge distillation for few-shot class-incremental learning[C] //2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2021: 2534-2543. DOI: 10.1109/CVPR46437.2021.00256.
[17] LAN T Z, ZENG F X, YI Z, et al.ICNoduleNet: enhancing pulmonary nodule detection performance on sharp kernel CT imaging[J]. IEEE Journal of Biomedical and Health Informatics, 2024, 28(8): 4751-4760. DOI: 10.1109/JBHI.2024.3402186.
[18] JOCHER G, STOKEN A, BOROVEC J, et al. Ultralytics/yolov5: v3.1-bug fixes and performance improvements[EB/OL]. (2020-10-29)[2024-12-30]. https://zenodo.org/records/4154370. DOI: 10.5281/zenodo.4154370.
[19] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C] //2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Los Alamitos, CA: IEEE Computer Society, 2018: 7132-7141. DOI: 10.1109/CVPR.2018.00745.
[20] 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. DOI: 10.1109/TPAMI.2015.2389824.
[21] HE Y H, ZHANG X Y, SUN J. Channel pruning for accelerating very deep neural networks[C] //2017 IEEE International Conference on Computer Vision (ICCV). Los Alamitos, CA: IEEE Computer Society, 2017: 1398-1406. DOI: 10.1109/ICCV.2017.155.
[22] YANG T J, HOWARD A, CHEN B, et al.NetAdapt: platform-aware neural network adaptation for mobile applications[C] //Computer Vision-ECCV 2018: LNCS Volume 11214. Cham: Springer, 2018: 289-304. DOI: 10.1007/978-3-030-01249-6_18.
[23] LI S J, GONG W Y, WANG L, et al. Evolutionary multitaskingvia reinforcement learning[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2024, 8(1): 762-775. DOI: 10.1109/TETCI.2023.3281876.
[24] TANG S N, ZHU Y, YUAN S Q. Intelligent fault identification of hydraulic pump using deep adaptive normalized CNN and synchrosqueezed wavelet transform[J]. Reliability Engineering & System Safety, 2022, 224: 108560. DOI: 10.1016/j.ress.2022.108560.
[25] WU B C, DAI X L, ZHANG P Z, et al.FBNet: hardware-aware efficient ConvNet design via differentiable neural architecture search[C] //2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2019: 10726-10734. DOI: 10.1109/CVPR.2019.01099.
[26] DONG X Y, YANG Y. Network pruning via transformable architecture search[C] //Advances in Neural Information Processing Systems 32 (NeurIPS 2019), Red Hook, NY: Curran Associates Inc., 2020: 69.
[27] SETIO A A A, TRAVERSO A, DE BEL T, et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challenge[J]. Medical Image Analysis, 2017, 42: 1-13. DOI: 10.1016/j.media.2017.06.015.
[28] FANG D, JIANG H, CHEN W Y, et al. Pulmonary nodule detection on lung parenchyma images usinghyber-deep algorithm[J]. Heliyon, 2023, 9(7): e17599. DOI: 10.1016/j.heliyon.2023.e17599.
[29] 赵奎, 仇慧琪, 李旭, 等. 结合注意力和多路径融合的实时肺结节检测算法[J]. 计算机应用, 2024, 44(3): 945-952. DOI: 10.11772/j.issn.1001-9081.2023040424.
[30] QIN R X, WANG Z Z, JIANG L Y, et al. Fine-grained lung cancer classification from PET and CT images based on multidimensional attention mechanism[J]. Complexity, 2020, 2020: 6153657. DOI: 10.1155/2020/6153657.
[31] 周涛, 叶鑫宇, 赵雅楠, 等. 跨模态注意力YOLOv5的PET/CT肺部肿瘤检测[J]. 中国图象图形学报, 2024, 29(4): 1070-1084. DOI: 10.11834/jig.230169.
[32] 张琼, 杭益柳, 邱建林, 等. 高效注意力金字塔网络在肺结节检测的应用[J]. 中国医学物理学杂志, 2024, 41(11): 1361-1369. DOI: 10.3969/j.issn.1005-202X.2024.11.007.
[33] JACOB C, MENON G C. Pathological categorization of lung carcinoma from multimodality images using convolutional neural networks[J]. International Journal of Imaging Systems and Technology, 2022, 32(5): 1681-1695. DOI: 10.1002/ima.22684.
[34] BARBOUCHI K, EL HAMDI D, ELOUEDI I, et al. A transformer-based deep neural network for detection and classification of lung cancer via PET/CT images[J]. International Journal of Imaging Systems and Technology, 2023, 33(4): 1383-1395. DOI: 10.1002/ima.22858.
[35] 梁智欣, 邓戈龙, 魏夏平, 等. 长尾分布下的PET-CT肺癌图像的深度学习网络检测与定位探索[J]. 中国医学物理学杂志, 2022, 39(12): 1473-1484. DOI: 10.3969/j.issn.1005-202X.2022.12.004.
[36] LI D H, LU Y, GAO Q, et al.LiteYOLO-ID: a lightweight object detection network for insulator defect detection[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 5023812. DOI: 10.1109/TIM.2024.3418082.
[1] 易见兵, 张裕贤, 曹锋, 李俊, 彭鑫, 陈鑫. 基于时空注意力的3D人体姿态估计网络设计[J]. 广西师范大学学报（自然科学版）, 2025, 43(5): 130-144.
[2] 魏明军, 周太宇, 纪占林, 张鑫楠. 基于YOLOv3的公共场所口罩佩戴检测方法[J]. 广西师范大学学报（自然科学版）, 2023, 41(1): 76-86.
[3] 张文龙, 南新元. 基于改进YOLOv5的道路车辆跟踪算法[J]. 广西师范大学学报（自然科学版）, 2022, 40(2): 49-57.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
[1] 刘晓娟, 林璐, 胡郁葱, 潘雷. 站点周边用地类型对地铁乘车满意度影响研究[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 1 -12 .
[2] 韩华彬, 高丙朋, 蔡鑫, 孙凯. 基于HO-CNN-BiLSTM-Transformer模型的风机叶片结冰故障诊断[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 13 -28 .
[3] 陈建国, 梁恩华, 宋学伟, 覃章荣. 基于OCT图像三维重建的人眼房水动力学LBM模拟[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 29 -41 .
[4] 李好, 何冰. 凹槽结构表面液滴弹跳行为研究[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 42 -53 .
[5] 凌福, 张永刚, 闻炳海. 基于插值的多相流格子Boltzmann方法曲线边界算法研究[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 54 -68 .
[6] 解盛, 马海菲, 张灿龙, 王智文, 韦春荣. 基于多分辨率特征定位的跨模态行人检索方法[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 69 -79 .
[7] 魏梓书, 陈志刚, 王衍学, 哈斯铁尔·马德提汗. 基于SBSI-YOLO11的轻量化轴承外观缺陷检测算法[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 80 -91 .
[8] 卢梦筱, 张阳春, 章晓峰. 基于分布式强化学习方法解决后继特征中的低估问题[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 107 -119 .
[9] 姜云卢, 卢辉杰, 黄晓雯. 惩罚加权复合分位数回归方法在固定效应面板数据中的应用研究[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 120 -127 .
[10] 邓金娜, 刘秋梅, 陈一鸣, 杨爱民. 两种黏弹性运动板的数值模拟与稳定性分析[J]. 广西师范大学学报（自然科学版）, 2025, 43(6): 128 -139 .
版权所有 © 广西师范大学学报（自然科学版）编辑部
地址：广西桂林市三里店育才路15号　邮编：541004
电话：0773-5857325　E-mail: gxsdzkb@mailbox.gxnu.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发