Related Experiment Video
Updated: Jan 13, 2026
06:34
SCAnED - An Open-source Skin Segmentation Macro for Semi-automated Cell and Nuclei Detection in Epidermal and Dermal Skin Compartments
Published on: August 8, 2025
519
BAF-UNet: a boundary-aware segmentation model for skin lesion segmentation
Menglei Zhang1, Congwei Zhang1, Zhibin Quan2
1Southeast University, School of Automation, Nanjing, China.
Journal of Medical Imaging (Bellingham, Wash.)
|January 12, 2026
Summary
We developed BAF-UNet, a novel deep learning model for precise skin lesion segmentation. This approach enhances boundary detection, improving accuracy in diagnosing skin cancer from medical images.
Area of Science:
- Medical image analysis
- Computer vision
- Dermatology
Background:
- Accurate skin lesion segmentation is crucial for effective skin cancer diagnosis and treatment.
- Challenges include ambiguous lesion boundaries, diverse shapes, and varying sizes, hindering precise segmentation.
Purpose of the Study:
- To enhance skin lesion segmentation performance by improving boundary preservation.
- To develop a deep learning model capable of accurately delineating skin lesion edges.
Main Methods:
- Proposed BAF-UNet, a boundary-aware segmentation network integrating multiscale boundary-aware feature fusion (BFF) and a boundary-aware vision transformer (BAViT).
- BAViT incorporates boundary guidance into MobileViT for capturing local and global context.
- Utilized a boundary-focused loss function to prioritize edge accuracy during training and evaluated on ISIC2016, ISIC2017, and PH2 datasets.
Main Results:
- BAF-UNet demonstrated improved Dice scores and boundary accuracy compared to baseline models.
- The BFF and BAViT modules effectively enhanced boundary delineation.
- The model maintained robustness across lesions of varying shapes and sizes.
Conclusions:
- BAF-UNet successfully integrates boundary guidance into feature fusion and transformer-based context modeling.
- Achieved significant improvements in segmentation accuracy, especially along lesion edges.
- Shows strong potential for clinical application in automated skin cancer diagnosis.
