International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Swin-RSCBNet: A Transformer-Based Network for Skin Cancer Segmentation with Multi-Scale and Attention Modules

Authors :

Benyamin Mirab Golkhatmi¹ Mostafa Heydari² Mahboobeh Houshmand³ Seyyed Abed Hosseini⁴

1- Department of Computer Engineering, Ma.C., Islamic Azad University, Mashhad, Iran. 2- Department of Computer Engineering, Ma.C., Islamic Azad University, Mashhad, Iran. 3- Department of Computer Engineering, Ma.C., Islamic Azad University, Mashhad, Iran. 4- Department of Electrical Engineering, Ma.C., Islamic Azad University, Mashhad, Iran.

Keywords :

Skin Cancer،Segmentation،Swin Transformer،Convolutional Block Attention Module،Receptive Field Block،Squeeze-and-Excitation

Abstract :

Skin lesion segmentation is critical in early skin cancer detection. However, achieving high segmentation accuracy while maintaining a lightweight model remains challenging. Here, we propose a novel architecture based on the Swin transformer called Swin-RSCBNet. The Swin transformer was selected as the backbone because it uses self-attention, which helps capture long-range dependencies necessary for accurate boundary detection. To further boost the model’s performance, we introduce a custom RSCB block, which combines three powerful modules: the convolutional block attention module (CBAM) to improve focus on salient spatial and channel features, the receptive field block (RFB) to capture multi-scale contextual information, and the squeeze-and-excitation (SE) block to enhance channel-wise feature recalibration. Swin-RSCBNet achieves a Dice score of 96.69% and an IoU of 93.59% on the PH2 dataset, a Dice score of 91.75%, and an IoU of 84.75% on the ISIC 2018 dataset. Swin-RSCBNet, leveraging the Swin transformer and a novel RSCB block, achieves superior skin lesion segmentation accuracy with a lightweight design for early skin cancer detection. Code available on Github https://github.com/Benyaminmirab/Swin-Segmention.git.