Boundary-Aware Multi-Stage with Mobile U-Net for Hair Segmentation in Dermoscopic Images
February 2026
in “
International journal of intelligent engineering and systems
”
TLDR The new method improves hair segmentation in skin images, helping detect skin cancer more accurately.
This study introduces a novel hair segmentation method for dermoscopic images, crucial for accurate skin lesion analysis and early cancer detection. The approach enhances the Mobile U-Net model by integrating multi-stage fine-tuning with boundary hybrid loss and incorporating CBAM or BGFF into skip connections. Tested on the ISIC image dataset, the method improved the Dice score from 76.97% to 79.08%, outperforming existing techniques. This advancement aids in precise artifact removal, supporting dermatologists and diagnostic tools in more reliably identifying skin cancer.
