Related Experiment Video
Updated: Sep 11, 2025
06:08
Quantitative Visualization and Detection of Skin Cancer Using Dynamic Thermal Imaging
Published on: May 5, 2011
16.9K
SGNet: A Structure-Guided Network with Dual-Domain Boundary Enhancement and Semantic Fusion for Skin Lesion
Haijiao Yun1, Qingyu Du1,2, Ziqing Han1,2
1School of Electronic Information Engineering, Changchun University, Changchun 130022, China.
Sensors (Basel, Switzerland)
|August 14, 2025
Summary
SGNet, a novel structure-guided network, enhances skin lesion segmentation using a hybrid CNN-Mamba framework. This approach improves accuracy in diagnosing skin cancers like melanoma by refining boundaries and integrating features effectively.
Area of Science:
- Dermatology and Medical Imaging
- Artificial Intelligence in Healthcare
- Computer Vision
Background:
- Accurate skin lesion segmentation is crucial for diagnosing skin cancers, including melanoma.
- Challenges include irregular shapes, blurred boundaries, low contrast, and artifacts in dermoscopic images.
- Existing deep learning models (UNet, Transformer) have limitations in feature extraction and computational efficiency.
Purpose of the Study:
- To introduce SGNet, a structure-guided network for robust skin lesion segmentation.
- To overcome limitations of conventional methods in feature exploitation and computational cost.
- To improve the precision of skin lesion delineation for computer-aided diagnosis.
Main Methods:
- Proposed SGNet integrates a hybrid Convolutional Neural Network (CNN)-Mamba framework.
- Utilized Visual Mamba (VMamba) encoder for efficient multi-scale feature extraction.
- Incorporated Dual-Domain Boundary Enhancer (DDBE), Semantic-Texture Fusion Unit (STFU), Structure-Aware Guidance Module (SAGM), and Guided Multi-Scale Refiner (GMSR).
Main Results:
- SGNet demonstrated superior performance on ISIC2017, ISIC2018, and PH2 datasets.
- Achieved average improvements of 3.30% in mean Intersection over Union (mIoU) and 1.77% in Dice Similarity Coefficient (DSC).
- Ablation studies confirmed the effectiveness of individual components and the overall network architecture.
Conclusions:
- SGNet offers exceptional accuracy and robust generalization for skin lesion segmentation.
- The hybrid CNN-Mamba approach effectively addresses challenges in dermoscopic image analysis.
- SGNet shows significant potential for advancing computer-aided dermatological diagnosis.
