TY  - JOUR
AU  - Mohammed, Essa
AU  - Khan, Abdullah
AU  - Ullah, Waqas
AU  - Khan, Wisal
AU  - Ahmed, Muhammad Jamal
PY  - 2025
DA  - 2025/06/05
TI  - Efficient Polyp Segmentation via Attention-Guided Lightweight Network with Progressive Multi-Scale Fusion
JO  - ICCK Transactions on Intelligent Systematics
T2  - ICCK Transactions on Intelligent Systematics
JF  - ICCK Transactions on Intelligent Systematics
VL  - 2
IS  - 2
SP  - 95
EP  - 108
DO  - 10.62762/TIS.2025.389995
UR  - https://www.icck.org/article/abs/TIS.2025.389995
KW  - colorectal cancer
KW  - visual intelligence
KW  - polyp segmentation
KW  - lightweight network
KW  - dual attention
KW  - multi-scale fusion
KW  - medical imaging
AB  - Accurate and real-time polyp segmentation plays a vital role in the early detection of colorectal cancer. However, existing methods often rely on computationally expensive backbones, single attention mechanisms, and suboptimal feature fusion strategies, limiting their practicality in real-world scenarios. In this work, we propose a lightweight yet effective deep learning framework that strikes a balance between precision and efficiency through a carefully designed architecture. Specifically, we adopt a MobileNetV4-based hybrid backbone to extract rich multi-scale features with significantly fewer parameters than conventional backbones, making the model well-suited for resource-constrained clinical settings. To enhance feature representation, we introduce a novel dual-attention guidance mechanism that integrates Efficient Channel Attention (ECA) for channel-wise refinement and Coordinate Attention (COA) for spatial modeling, which is particularly effective at delineating polyp boundaries. Additionally, we design a progressive multi-scale fusion strategy that hierarchically integrates feature maps from deep to shallow layers, preserving spatial details while enhancing contextual understanding. Extensive experiments on five benchmark polyp segmentation datasets demonstrate that our method consistently outperforms state-of-the-art approaches across both quantitative metrics and qualitative visualizations. Comprehensive ablation studies further validate the effectiveness of each component, highlighting the practical viability of our approach for real-time polyp segmentation applications.
SN  - 3068-5079
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Mohammed2025Efficient,
  author = {Essa Mohammed and Abdullah Khan and Waqas Ullah and Wisal Khan and Muhammad Jamal Ahmed},
  title = {Efficient Polyp Segmentation via Attention-Guided Lightweight Network with Progressive Multi-Scale Fusion},
  journal = {ICCK Transactions on Intelligent Systematics},
  year = {2025},
  volume = {2},
  number = {2},
  pages = {95-108},
  doi = {10.62762/TIS.2025.389995},
  url = {https://www.icck.org/article/abs/TIS.2025.389995},
  abstract = {Accurate and real-time polyp segmentation plays a vital role in the early detection of colorectal cancer. However, existing methods often rely on computationally expensive backbones, single attention mechanisms, and suboptimal feature fusion strategies, limiting their practicality in real-world scenarios. In this work, we propose a lightweight yet effective deep learning framework that strikes a balance between precision and efficiency through a carefully designed architecture. Specifically, we adopt a MobileNetV4-based hybrid backbone to extract rich multi-scale features with significantly fewer parameters than conventional backbones, making the model well-suited for resource-constrained clinical settings. To enhance feature representation, we introduce a novel dual-attention guidance mechanism that integrates Efficient Channel Attention (ECA) for channel-wise refinement and Coordinate Attention (COA) for spatial modeling, which is particularly effective at delineating polyp boundaries. Additionally, we design a progressive multi-scale fusion strategy that hierarchically integrates feature maps from deep to shallow layers, preserving spatial details while enhancing contextual understanding. Extensive experiments on five benchmark polyp segmentation datasets demonstrate that our method consistently outperforms state-of-the-art approaches across both quantitative metrics and qualitative visualizations. Comprehensive ablation studies further validate the effectiveness of each component, highlighting the practical viability of our approach for real-time polyp segmentation applications.},
  keywords = {colorectal cancer, visual intelligence, polyp segmentation, lightweight network, dual attention, multi-scale fusion, medical imaging},
  issn = {3068-5079},
  publisher = {Institute of Central Computation and Knowledge}
}