Precise segmentation of colon polyps is crucial for early screening and clinical diagnosis of colorectal cancer. However, large variations in polyp size, shape, and boundary characteristics, along with low contrast and blurred edges, make it challenging to balance segmentation accuracy and computational efficiency. To address these issues, a lightweight dynamic multi-scale segmentation network (LDMS-Net) was proposed. The model adopted a U-shaped encoder-decoder architecture and introduced a dynamic multi-scale module (DMM) in the encoder. Specifically, global average pooling and a two-layer fully connected network were used to generate adaptive weights, enabling dynamic fusion of multiple convolution branches with different dilation rates (1,2, 3) and a standard convolution branch for adaptive multi-scale feature representation. A structural re-parameterization strategy was further applied to merge the multi-branch structure into a single convolution during inference to reduce computational cost. In addition, a regional detail aggregator (RDA) and a comprehensive context integrator (CCI) were designed to capture local texture details and global semantic dependencies, respectively, and were integrated in the decoder via a hierarchical late fusion strategy. Experiments on Kvasir-SEG and CVC-ClinicDB datasets show that, using only 800 training images from Kvasir-SEG, the proposed method achieves a mean Dice coefficient of 93.65% and a mean IoU of 89.39%, with 6.54×106 parameters and 7.23×109 floating-point operations. The results demonstrate that LDMS-Net achieves high accuracy and robustness while maintaining a lightweight design.
LITJENSG, KOOIT, BEJNORDIB E, et al. A survey on deep learning in medical image analysis[J]. Medical Image Analysis, 2017, 42: 60-88.
[2]
DIJ, ZHUY, LIANGC. A medical image segmentation model based on SAM with an integrated local multiscale feature encoder[J]. Journal of Measurement Science and Instrumentation, 2025, 16(3): 359-370.
[3]
NIIKURAR, HIRATAY, SUZUKIN, et al. Colonoscopy reduces colorectal cancer mortality: A multicenter, long-term, colonoscopy-based cohort study[J]. PLoS One, 2017, 12(9): e0185294.
[4]
BERNALJ, SÁNCHEZF J, FERNÁNDEZ-ESPARRACHG, et al. WM-DOVA maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians[J]. Computerized Medical Imaging and Graphics, 2015, 43: 99-111.
[5]
AMELINGS, WIRTHS, PAULUSD, et al. Texture-based polyp detection in colonoscopy[C]//Bildverarbeitung für die Medizin 2009. Berlin, Heidelberg: Springer, 2009: 346-350.
[6]
BERNALJ, SÁNCHEZJ, VILARIÑOF. Towards automatic polyp detection with a polyp appearance model[J]. Pattern Recognition, 2012, 45(9): 3166-3182.
[7]
HWANGS, OH J, TAVANAPONGW, et al. Polyp detection in colonoscopy video using elliptical shape feature[C]//2007 IEEE International Conference on Image Processing, 2007: II-465-II-468.
[8]
TAJBAKHSHN, SHINJ Y, GURUDUS R, et al. Convolutional neural networks for medical image analysis: Full training or fine tuning?[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1299-1312.
[9]
RONNEBERGERO, FISCHERP, BROXT. U-Net: Convolutional networks for biomedical image segmentation[C]//Medical Image Computing and Computer-Assisted Intervention—MICCAI 2015. Cham: Springer, 2015: 234-241.
[10]
ZHOUZ, RAHMAN SIDDIQUEEM M, TAJBAKHSHN, et al. UNet++: A nested U-Net architecture for medical image segmentation[C]//Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. Cham: Springer, 2018: 3-11.
[11]
ZHANGZ, LIUQ, WANGY. Road extraction by deep residual U-net[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(5): 749-753.
[12]
OKTAYO, SCHLEMPERJ, LE FOLGOCL, et al. Attention U-Net: Learning where to look for the pancreas[DB/OL]. (2018-05-20)[2025-09-25].
[13]
WANGX, GIRSHICKR, GUPTAA, et al. Non-local neural networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 7794-7803.
[14]
DOSOVITSKIYA, BEYERL, KOLESNIKOVA, et al. An image is worth 16×16 words: Transformers for image recognition at scale[DB/OL]. (2021-06-03)[2025-09-25].
[15]
CHENJ, LUY, YUQ, et al. TransUNet: Transformers make strong encoders for medical image segmentation[DB/OL]. (2021-06-03)[2025-09-25].
[16]
CAOH, WANGY, CHENJ, et al. Swin-Unet: Unet-like pure transformer for medical image segmentation[C]//Computer Vision-ECCV 2022 Workshops. Cham: Springer, 2023: 205-218.
[17]
VALANARASUJ M J, OZA P, HACIHALILOGLUI, et al. Medical transformer: Gated axial-attention for medical image segmentation[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2021. Cham: Springer, 2021: 36-46.
[18]
LINT Y, DOLLÁRP, GIRSHICKR, et al. Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017: 936-944.
[19]
CHENL C, PAPANDREOUG, SCHROFFF, et al. Rethinking atrous convolution for semantic image segmentation[DB/OL]. (2017-12-05)[2025-09-25].
[20]
YANGB, BENDERG, LEQ V, et al. CondConv: Conditionally parameterized convolutions for efficient inference[DB/OL]. (2017-12-05)[2025-09-25].
[21]
DINGX, ZHANGX, MAN, et al. RepVGG: Making VGG-style ConvNets great again[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021: 13728-13737.
[22]
JHA D, SMEDSRUDP H, RIEGLERM A, et al. Kvasir-SEG: A segmented polyp dataset[C]//MultiMedia Modeling. Cham: Springer, 2020: 451-462.
[23]
BERNALJ, SÁNCHEZJ, VILARIÑOF. Towards automatic polyp detection with a polyp appearance model[J]. Pattern Recognition, 2012, 45(9): 3166-3182.
[24]
BERNALJ, SÁNCHEZF J, FERNÁNDEZ-ESPARRACHG, et al. WM-DOVA maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians[J]. Computerized Medical Imaging and Graphics, 2015, 43: 99-111.
[25]
SILVAJ, HISTACEA, ROMAINO, et al. Toward embedded detection of polyps in WCE images for early diagnosis of colorectal cancer[J]. International Journal of Computer Assisted Radiology and Surgery, 2014, 9(2): 283-293.
[26]
WANGJ, HUANGQ, TANGF, et al. Stepwise feature fusion: Local guides global[C]//Medical Image Computing and Computer Assisted Intervention-MICCAI 2022. Cham: Springer, 2022: 110-120.
[27]
JHA D, SMEDSRUDP H, RIEGLERM A, et al. ResUNet++: An advanced architecture for medical image segmentation[C]//2019 IEEE International Symposium on Multimedia (ISM), 2020: 225-2255.
[28]
CHENL C, ZHUY, PAPANDREOUG, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Computer Vision-ECCV 2018. Cham: Springer, 2018: 833-851.
[29]
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[DB/OL]. (2021-01-20)[2025-09-25].
[30]
QINX, ZHANGZ, HUANGC, et al. U2-Net: Going deeper with nested U-structure for salient object detection[J]. Pattern Recognition, 2020, 106: 107404.