In order to solve the problems of failing to accurately distinguish polyps from the surrounding mucosa and the limited generalization ability in clinical polyp segmentation algorithms, a new polyp segmentation network that combines cross-layer feature fusion and boundary guidance was proposed. It primarily consists of three modules: the multiscale feature extraction module was used to extract multiscale information and polyp details from the encoder’s shallow output. The cross-layer feature fusion module was designed to integrate the cross-layer features through collaborative attention and adaptive channel calibration strategies, achieving dynamic calibration and fusion of these features. The boundary extraction guidance module involves the use of morphological operators for the extraction of boundaries and polyp regions from the initial segmentation map. Features within the internal polyp region were leveraged to enhance those near the boundary, resulting in accurate segmentation outcomes. The experimental results show that the model’s Dice coefficients on the CVC-ClinicDB and Kvasir datasets are 94.24% and 92.41%, respectively. Compared with the existing mainstream methods, the proposed method not only segments multiple types of polyps but also improves the segmentation accuracy and generalization ability.
SUN Jian-jun, CUI Yu-xia, JIA Jing-hao, LI Xue.
Polyp Image Segmentation Based on Boundary Guidance and Cross-layer Feature Fusion[J].
Science Technology and Engineering, 2026, 26(5): 2060-2067 DOI:10.12404/j.issn.1671-1815.2500862
为了验证模型在参数数量和计算资源方面的效率,在实验中使用了模型参数和每秒浮点运算次数(floating point operations per second,FLOPs)两个指标。FLOPs是模型计算能力的指标,较低的FLOPs值表明模型需要的计算资源较少。表5总结了各种基准方法中参数量和FLOPs的比较。结果表明,与Polyp-CASCADE相比,所提出的模型中的参数数量减少了28.8%。与其他基于Transformer结构的模型相比,所提出的模型具有最低的参数数量,并且FLOPs处于较低水平。这也意味着在推理和部署过程中需要更少的资源并且运行效率更高。
ZhangPanfeng, YangHe, ShenXianhao, et al. Polyp segmentation model based on fusion of local and global features[J]. Electronic Measurement Technology, 2024, 47(16): 100-109.
[5]
DumitruR G, PeteleazaD, CraciunC. Using DUCK-Net for polypimage segmentation[J]. Scientific Reports, 2023, 13(1): 9803.
[6]
NguyenT C, NguyenT P, DiepG H, et al. CCBANet: cascading context and balancing attention for polyp segmentation[C]// International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin: Springer, Cham, 2021: 633-643.
[7]
YinZ, LiangK, MaZ, et al. Duplex contextual relation network for polyp segmentation[C]// 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI). Kolkata: IEEE, 2022: 1-5.
LiQing, HuangfuYubin, LiJiangyun, et al. UConvTrans: a dual-flow cardiac image segmentation network by global and local information integration[J]. Journal of Shanghai Jiao Tong University, 2023, 57(5): 570-581.
LiDaxiang, LiDenghui, LiuYing, et al. Progressive CNN-transformer semantic compensation network for polyp segmentation[J]. Optics and Precision Engineering, 2024, 32(16): 2523-2536.
[12]
DosovitskiyA. An image is worth 16×16 words: transformers for image recognition at scale[J]. ArXiv Preprint, ArXiv: 2010.11929, 2020.
[13]
FuB, PengY, HeJ, et al. HmsU-Net: a hybrid multi-scale U-net based on a CNN and transformer for medical image segmentation[J]. Computers in Biology and Medicine, 2024, 170: 108013.
[14]
WuY H, LiuY, ZhanX, et al. P2T: pyramid pooling transformer for scene understanding[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(11): 12760-12771.
[15]
DuH, WangJ, LiuM, et al. SwinPA-Net: swin transformer-based multiscale feature pyramid aggregation network for medical image segmentation[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 35(4): 5355-5366.
WangTiantian, ShiWeiya, ZhangShiqiang, et al. Optic disk and cup segmentation using dual brand and transformer[J]. Science Technology and Engineering, 2023, 23(6): 2499-2508.
DingWeilong, ZongZeyong, ZhuHongbo, et al. Histopathological image segmentation method based on CNN and Transformer[J]. Zhejiang University of Technology, 2024, 52(6): 591-600.
[20]
WeiJ, HuY, ZhangR, et al. Shallow attention network for polyp segmentation[C]// Medical Image Computing and Computer Assisted Intervention-MICCAI 2021: 24th International Conference. Strasbourg: Springer International Publishing, 2021: 699-708.
[21]
LiuG, YaoS, LiuD, et al. CAFE-Net: cross-attention and feature exploration network for polyp segmentation[J]. Expert Systems with Applications, 2024, 238: 121754.
[22]
XuZ, TianB, LiuS, et al. Collaborative attention guided multi-scale feature fusion network for medical image segmentation[J]. IEEE Transactions on Network Science and Engineering, 2023, 11(2): 1857-1871.
[23]
WeiJ, WangS, HuangQ. F3Net: fusion, feedback and focus for salient object detection[C]// Proceedings of the AAAI Conference on Artificial Intelligence. New York: AAAI, 2020, 34(7): 12321-12328.
[24]
FanD P, JiG P, ZhouT, et al. Pranet: parallel reverse attention network for polyp segmentation[C]// International Conference on Medical Image Computing and Computer-assisted Intervention. Cham: Springer International Publishing, 2020: 263-273.
[25]
HuangC H, WuH Y, LinY L. Hardnet-mseg: a simple encoder-decoder polyp segmentation neural network that achieves over 0.9 mean dice and 86 FPS[J]. ArXiv Preprint, ArXiv: 2101.07172, 2021.
[26]
HuangX, DengZ, LiD, et al. Missformer: an effective medical image segmentation transformer[J]. ArXiv Preprint, ArXiv: 2109.07162, 2021.
[27]
DongB, WangW, FanD P, et al. Polyp-PVT: polyp segmentation with pyramid vision transformers[J]. ArXiv, 2021: 2108.06932.
[28]
ZhaoX, JiaH, PangY, et al. Multi-scale in multi-scale subtraction network for medical image segmentation[J]. ArXiv Preprint, ArXiv: 2303.10894, 2023.
[29]
RahmanM M, MarculescuR. Medical image segmentation via cascaded attention decoding[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. Waikoloa: IEEE, 2023: 6222-6231.
PDF
(6770KB)
