.png){kind=icon}
.png){kind=icon}
Return
Skin lesion segmentation method based on lightweight context aware network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2026.114624.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In recent years, significant progress has been made in skin lesion segmentation methods based on deep learning. However, accurately distinguishing the boundaries of lesions from the regions of interest with a small parameter count and low computational complexity still remains challenging. Moreover, real-time performance is also crucial because the rapid acquirement of accurate segmentation results can assist medical professionals in making timely and correct decisions. This paper proposes a lightweight context-aware skin lesion segmentation network (LCS-Net) that features extremely low network complexity and short inference time. The proposed model integrates novel modules for efficient feature extraction, multi-scale context aggregation, and boundary refinement. Finally, we compare LCS-Net with several state-of-the-art methods on three publicly available datasets: skin lesion segmentation datasets provided by international skin imaging collaboration ISIC2017, ISIC2018 and a lung segmentation dataset. Experimental results demonstrate that LCS-Net achieves a Dice coefficient of 89.41% and a Jaccard Index of 80.86% on the ISIC2018 dataset, outperforming state-of-the-art methods such as U-Net (Dice: 87.55%, JI: 77.86%) and TransFuse (Dice: 89.27%, JI: 80.63%). With the parameter count of 0.07 million and the inference time of 6.8 ms, LCS-Net consistently outperforms other state-of-the-art networks in segmentation accuracy, computational efficiency, and model size, showing its application potential in resource-constrained.
Keywords:
skin lesion segmentation
lightweight network
context-aware
boundary refinement
computational efficiency
Journal
IF:
8
Papers:
5.4K
Citations:
3.5W
Organization
No organization information available