Related Experiment Video
Updated: May 21, 2025
06:08
Quantitative Visualization and Detection of Skin Cancer Using Dynamic Thermal Imaging
Published on: May 5, 2011
16.7K
GatedSegDiff: a gated fusion diffusion model for skin lesion segmentation
Rui Wang1, Liucheng Yao1, Jiawen Zeng1
1School of Communication and Information Engineering, Shanghai University, Shanghai, 200444, China.
Medical & Biological Engineering & Computing
|March 18, 2025
Summary
GatedSegDiff improves skin lesion segmentation accuracy, especially for fuzzy boundaries, aiding melanoma diagnosis. This deep learning model enhances diagnostic precision in digital dermatology.
Area of Science:
- Dermatology
- Medical Image Analysis
- Artificial Intelligence
Background:
- Accurate skin lesion segmentation is critical for diagnosing skin diseases like melanoma.
- Existing methods struggle with the fuzzy boundaries characteristic of many skin lesions.
- Advanced image segmentation techniques are needed to improve diagnostic precision.
Purpose of the Study:
- To develop an end-to-end framework, GatedSegDiff, for precise melanoma skin lesion segmentation.
- To address the challenge of segmenting skin lesions with unclear boundaries.
- To enhance the accuracy and efficiency of skin lesion analysis.
Main Methods:
- Developed GatedSegDiff, an end-to-end framework integrating denoising networks and a gated attention fusion module.
- Utilized semantic representation and multi-scale feature map merging for enhanced segmentation.
- Evaluated the model on ISIC 2017, ISIC 2018, and PH2 skin lesion image datasets.
Main Results:
- GatedSegDiff achieved an average 4.3% increase in IoU score across three datasets.
- The model demonstrated a 1.5% decrease in HD95 score, indicating improved boundary accuracy.
- Outperformed existing advanced methods in multiple performance metrics for skin lesion segmentation.
Conclusions:
- GatedSegDiff significantly advances skin lesion segmentation, particularly for lesions with fuzzy boundaries.
- The model enhances diagnostic precision and efficiency in digital dermatological applications.
- Shows potential for clinical use in early skin disease diagnosis and treatment planning.
