面向CT肺结节分割的动态区域优化方法

杨国虎; 刘平; 金雨纯; 甄元元

doi:null

您当前的位置：

首页 >

文章列表页 >

面向CT肺结节分割的动态区域优化方法

更新时间：2026-03-03

- 面向CT肺结节分割的动态区域优化方法
- Dynamic Region Optimization Method for CT Pulmonary Nodule Segmentation
- 物联网学报 2026年
- 作者机构：
  
  1.宁夏大学电子与电气工程学院，宁夏银川 750021
  2.宁夏大学宁夏沙漠信息智能感知重点实验室，宁夏银川 750021
- 作者简介：
  
  E-mail：Liuping@nxu.edu.cn
- 基金信息：
  
  宁夏自然科学基金(2023AAC03140)
- DOI：
  中图分类号：
- 收稿：2025-08-05，
  
  修回：2025-09-10，
  
  录用：2025-10-20，
- 稿件说明：
移动端阅览
杨国虎, 刘平, 金雨纯, 等. 面向CT肺结节分割的动态区域优化方法[J/OL]. 物联网学报, 2026.

YANG Guohu, LIU Ping, JIN Yuchun, et al. Dynamic Region Optimization Method for CT Pulmonary Nodule Segmentation[J/OL]. Chinese Journal on Internet of Things, 2026.
杨国虎, 刘平, 金雨纯, 等. 面向CT肺结节分割的动态区域优化方法[J/OL]. 物联网学报, 2026. DOI：

YANG Guohu, LIU Ping, JIN Yuchun, et al. Dynamic Region Optimization Method for CT Pulmonary Nodule Segmentation[J/OL]. Chinese Journal on Internet of Things, 2026. DOI：

摘要

肺结节的精准分割是肺癌早期筛查与临床诊断的关键前提，然而 CT 影像中结节存在形态不规则、边界模糊、尺寸差异大（从几毫米到数十毫米）等问题，且易受血管、胸膜等周围组织干扰，导致传统深度学习分割方法（如 U-Net、基线 V-Net）存在边界定位误差大、小病灶漏诊率高、分割一致性差等局限；此外，固定感兴趣区域（ROI）的设置易造成目标信息冗余或关键细节丢失，进一步制约分割精度。针对上述挑战，本研究提出一种两阶段肺结节分割方法，结合自适应ROI算法与多视角三维分割策略。第一阶段采用V-Net架构沿轴向进行初始分割，通过创新的 A-ROI 算法动态调整ROI位置和尺寸，保持结节与ROI面积比低于阈值 RT（经实验确定为 0.6），减少无关组织干扰；第二阶段沿冠状和矢状轴进行补丁式分析，最终通过共识模块整合多平面预测结果（一致性比率设为 50%），提升分割稳定性。在 LUNA16和LNDb公开数据集上的实验表明，该方法Dice 系数分别达92.6%和92.3%，较基线V-Net提升6.2和6.1个百分点，Hausdorff 距离降低至 2.92±1.89mm；相较于传统 U-Net，分割精度亦有显著提升。消融实验验证：自适应 ROI 使边界误差减少 37.5%，多平面协同分析提升形状相似度 29.8%，能有效解决CT 影像中结节分割的核心挑战，为临床肺癌早期精准诊断与疗效评估提供可靠技术支持。

Abstract

Accurate segmentation of pulmonary nodules is a crucial prerequisite for early lung cancer screening and clinical diagnosis. However

pulmonary nodules in CT images exhibit challenges such as irregular shapes

blurred boundaries

and significant size variations (ranging from a few millimeters to tens of millimeters). Additionally

they are easily interfered with by surrounding tissues like blood vessels and pleura

which leads to limitations in traditional deep learning segmentation methods (e.g.

U-Net

baseline V-Net)

including large boundary localization errors

high missed diagnosis rates of small lesions

and poor segmentation consistency. Furthermore

the fixed setting of the region of interest (ROI) tends to cause redundancy of target information or loss of key details

further restricting segmentation accuracy.To address the above challenges

this study proposes a two-stage pulmonary nodule segmentation method that combines an adaptive ROI algorithm with a multi-view 3D segmentation strategy. In the first stage

the V-Net architecture is employed to perform initial segmentation along the axial axis. An innovative adaptive ROI (A-ROI) algorithm dynamically adjusts the position and size of the ROI

maintaining the area ratio of the nodule to the ROI below a threshold RT (experimentally determined as 0.6) to reduce interference from irrelevant tissues. In the second stage

patch-based analysis is conducted along the coronal and sagittal axes

and finally

a consensus module integrates the multi-plane prediction results (with a consistency ratio set to 50%) to enhance segmentation stability.Experiments on the public LUNA16 and LNDb datasets show that this method achieves Dice coefficients of 92.6% and 92.3%

respectively

representing improvements of 6.2 and 6.1 percentage points compared to the baseline V-Net

while the Hausdorff distance is reduced to 2.92±1.89 mm. Compared with the traditional U-Net

the segmentation accuracy is also significantly improved. Ablation experiments verify that the adaptive ROI reduces boundary errors by 37.5%

and multi-plane collaborative analysis improves shape similarity by 29.8%. This method effectively addresses the core challenges in pulmonary nodule segmentation in CT images

providing reliable technical support for clinical accurate diagnosis of early lung cancer and efficacy evaluation.

关键词

Keywords

references

Zhong Ruikang , Li Jinghua , Lin Ximing , et al . Progress in the Application of Deep Learning in Prognostic Models for Non-Small Cell Lung Cancer. Chinese Journal of Thoracic and Cardiovascular Surgery, 2024, 31(09): 1345-1350.钟睿康,李靖华,林熙明, 等. 深度学习在非小细胞肺癌预后模型中的应用进展 [J ] . 中国胸心血管外科临床杂志 , 2024 , 31 ( 09 ): 1345 - 1350 .

Xie Juanying , Zhang Kaiyun . SOSNet: An Asymmetric Encoder-Decoder Structure for CT Image Segmentation of Non-Small Cell Lung Cancer . Electronic Journal , 2024 , 52 ( 03 ): 824 - 837 .

谢娟英 , 张凯云 . SOSNet:一种非对称编码器-解码器结构的非小细胞肺癌CT图像分割模型 [J ] . 电子学报 , 2024 , 52 ( 03 ): 824 - 837 .

Usman M. , Latif S. , Asim M. , Lee B.-D. & Qadir, J. Retrospective motion correction in multishot mri using generative adversarialnetwork . Sci. Rep . 10 , 1 – 11 ( 2020 ).

LIANGLM , ZHOULS , YUJ , etal . MultiscalefeaturefusiondoubleU-shapedretinalsegmentationalgorithm [J ] . JournalofOptoelectronics·Laser , 2022 , 33 ( 3 ): 272

梁礼明 , 周珑颂 , 余洁 , 等 . 多尺度特征融合双U型视网膜分割算法 [J ] . 光电子·激光 , 2022 , 33 ( 3 ): 272 - 282 .

ALOMMZ , HASANM , YAKOPCICC , etal . Recurrent residualconvolutionalneuralnetwork based onU-Net(R2U-Net)formedicalimagesegmentation [EB/OL ] .( 2018-05-29 )[ 2022-03-02 ] . https://arxiv.org/abs/ 1802.06955 https://arxiv.org/abs/1802.06955 .

KUCHARSKID , KLECZEKP , KORJAKOWSKAJJ , etal . Semi-supervised nests of melanocytes segmentationmethodusingconvolutionalautoencoders [J ] . Sensors , 2020 , 20 ( 6 ): 1546 - 1565 .

WANG Z R , MEN J R , ZHANG F C . Improved V-Net lung nodule segmentation method based on selective kernel [J ] . Signal, Image and Video Processing , 2023 , 17 ( 5 ): 1763 - 1774 .

Chen Y , Zhang Y , Zhang X , et al . Characterization of adrenal glands on computed tomography with a 3D V-Net-based model . [J ] . Insights into imaging , 2025 , 16 ( 1 ): 17 .

Xiao Hongyu , Yang Weidong , Wang Qi . " A Deep Learning-Based Algorithm for Dual-Phase CT Liver Cancer Detection " [J/OL ] . Optoelectronics and Lasers , 1 -10 [April 16, 2025 ] . http://kns.cnki.net/kcms/detail/12.1182.O4.20240428.1411.006.html http://kns.cnki.net/kcms/detail/12.1182.O4.20240428.1411.006.html .

肖宏宇 , 杨伟东 , 王琦 . 基于深度学习的双期相CT肝癌检测算法 [J/OL ] . 光电子·激光 , 1 - 10 [ 2025-04-16 ] .

Hang L , Shaowei C , Xiaohui Z , et al . Detection of obstructive sleep apnea from single-channel ECG signals using a CNN-transformer architecture [J ] . Biomedical Signal Processing and Control , 2023 , 82 .

卢娟 , 汤铁群 , 张荣福 , 等 . 融合多特征和注意力机制的肺结节分割算法 [J ] . 小型微型计算机系统 , 2023 , 44 ( 01 ): 43 - 48 . DOI: 10.20009/j.cnki.21-1106/TP.2021-0378 http://dx.doi.org/10.20009/j.cnki.21-1106/TP.2021-0378 .

LI Q H , ZHOU X F , XU Y S . Research on Medical Multi-CT Image Encryption Algorithm Based on DNA Scrambling and Optical Chaos [J/OL ] . Journal of Optoelectronics·Laser , 1 - 12 [ 2025-03-22 ] . http://kns.cnki.net/kcms/detail/12.1182.O4.20241009.1556.012.html http://kns.cnki.net/kcms/detail/12.1182.O4.20241009.1556.012.html .

李奇翰 , 周雪芳 , 徐远胜 . 基于DNA置乱和光混沌医学多 CT图像加密算法研究 [J/OL ] . 光电子·激光 , 1 - 12 [ 2025-03-22 ] . http://kns.cnki.net/kcms/detail/12.1182.O4.20241009.1556.012.html http://kns.cnki.net/kcms/detail/12.1182.O4.20241009.1556.012.html .

WANG B X , SONG Y S , DONG X N . Difficulties in Attention Perception for Infrared Small Target Detection Network [J ] . Chinese Optics (in Chinese and English) , 2024 , 17 ( 03 ): 538 - 547 .

王伯霄 , 宋延嵩 , 董小娜 . 难点注意力感知红外小目标检测网络 [J ] . 中国光学(中英文) , 2024 , 17 ( 03 ): 538 - 547 .

ZHU X , WANG X , SHI Y , et al . Channel-wise attention mechanism in the 3d convolutional network for lung nodule detection [J ] . Electronics , 2022 , 11 ( 10 ): 1600 .

Zhu Hegui , Dai Lewen , Liu Yating , et al . A three-dimensional bit-level image encryption algorithm with Rubik's cube method [J ] . Mathematics and Computers in Simulation , 2021 , 185 : 754 - 770 .

Shuvo M. B. , Ahommed R. , Reza S. & Hashem M. M. A. CNL-UNet: A novel lightweight deep learning architecture for multi-modal biomedical image segmentation with false output suppression . Biomed. Signal Process. Control 70 , 102959 . https:// doi. org/10. 1016/j. bspc. 2021. 102959 (2021).

Lama N. , Hagerty J. , Nambisan A. , Stanley R. J. & Van Stoecker W. Skin lesion segmentation in dermoscopic images with noisydata . J. Dig. Imaging https:// doi. org/ 10. 1007/ s10278- 023- 00819-8 (2023) .

Upadhyay S. , Beevi A.S. & Kalady S. Left Ventricle segmentation of 2D Echocardiography using deep learning . In InternationalConference on Computer Vision and Image Processing 87 – 98 (Springer, 2022 ). https:// doi. org/ 10. 1007/ 978-3- 031- 31407-0_7.

Xiaotian L , Ailian J , Yanfang Q , et al . TPFR-Net: U-shaped model for lung nodule segmentation based on transformer pooling and dual-attention feature reorganization . [J ] . Medical & biological engineering & computing , 2023 , 61 ( 8 ): 1929 - 1946 .

Carles M , Kuhn D , Fechter T , et al . Correction: Development and evaluation of two open-source nnU-Net models for automatic segmentation of lung tumors on PET and CT images with and without respiratory motion compensation . [J ] . European radiology , 2024 , 34 ( 12 ): 8039 - 8039 .

Lafraxo S , El Ansari M , Koutti L , et al . SEDARU-net: a squeeze-excitation dilated based residual U-Net with attention mechanism for automatic melanoma lesion segmentation [J ] . Multimedia Tools and Applications , 2024 , 84 ( 21 ): 1 - 28 .

Palliyil S S , Rohini P , Ramakrishnan S . An Approach to Segment Nuclei and Cytoplasm in Lung Cancer Brightfield Images Using Hybrid Swin-Unet Transformer [J ] . Journal of Medical and Biological Engineering , 2024 , 44 ( 3 ): 448 - 459 .

Sweetline C B , Vijayakumaran C , Samydurai A . Overcoming the Challenge of Accurate Segmentation of Lung Nodules: A Multi-crop CNN Approach . [J ] . Journal of imaging informatics in medicine , 2024 , 37 ( 3 ): 988 - 1007 .

Zhu L , Wang Y , Huang A , et al . A multi-branch, multi-scale, and multi-view CNN with lightweight temporal attention mechanism for EEG-based motor imagery decoding . [J ] . Computer methods in biomechanics and biomedical engineering , 2025 , 11 - 15 .

Lv Z , Xu B , Zhong L , et al . Improved monitoring of southern corn rust using UAV-based multi-view imagery and an attention-based deep learning method [J ] . Computers and Electronics in Agriculture , 2024 , 224 109232 - 109232 .

Ben A S , Hanene S , Yessine A , et al . Res-Net-VGG19: Improved tumor segmentation using MR images based on Res-Net architecture and efficient VGG gliomas grading [J ] . Applications in Engineering Science , 2023 , 16 .

Expert Systems ; Reports Outline Expert Systems Findings from E . Shakibapour and Colleagues (An Unsupervised Metaheuristic Search Approach for Segmentation and Volume Measurement of Pulmonary Nodules In Lung Ct Scans) [J ] . Journal of Robotics & Machine Learning , 2019 .

Muhammad U , Byoung-Dai L , Shi-Sub B , et al . Volumetric lung nodule segmentation using adaptive ROI with multi-view residual learning . [J ] . Scientific reports , 2020 , 10 ( 1 ): 12839 .

J J , Haw C S , Palanichamy N , et al . IM- LTS: An Integrated Model for Lung Tumor Segmentation using Neural Networks and IoMT [J ] . MethodsX , 2025 , 14 103201 - 103201 .

Minglei L , Hang Z , Xiang L , et al . SDA-Net: Self-distillation driven deformable attentive aggregation network for thyroid nodule identification in ultrasound images [J ] . Artificial Intelligence In Medicine , 2023 , 146 102699 - 102699 .

Liu J , Suo G , Jin F , et al . WGAN-based multi-structure segmentation of vertebral cross-section MRI using ResU-Net and clustered transformer [J ] . Scientific Reports , 2024 , 14 ( 1 ): 27474 - 27474 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据