Swin-WaveNet: frequency-guided boundary restoration for robust skin lesion segmentation.
8 affiliations
- 1Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, China Three Gorges University, Yichang, Hubei, 443002, China.
- 2College of Computer and Information Technology, China Three Gorges University, Yichang, Hubei, 443002, China.
- 3Yichang Key Laboratory of Intelligent Medicine, China Three Gorges University, Yichang, Hubei, 443002, China.
- 4Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering, China Three Gorges University, Yichang, Hubei, 443002, China. [email protected].
- 5College of Computer and Information Technology, China Three Gorges University, Yichang, Hubei, 443002, China. [email protected].
- 6Yichang Key Laboratory of Intelligent Medicine, China Three Gorges University, Yichang, Hubei, 443002, China. [email protected].
- 7Institute of Advanced Studies in Humanities and Social Sciences, Beijing Normal University, Zhuhai, Guangdong, 519087, China.
- 8School of Information Science and Technology, Hangzhou Normal University, Hangzhou, Zhejiang, 311121, China.
Abstract
Accurate skin lesion segmentation remains difficult when dermoscopic boundaries are fragmented by hair occlusion, uneven illumination, or weak local contrast. To address this problem, the authors propose Swin-WaveNet, a frequency-guided encoder-decoder that combines a Swin Transformer backbone with wavelet decomposition and selective state-space modeling. Rather than passing skip features directly to the decoder, the proposed method separates low-frequency semantic information from high-frequency details, selectively refines boundary-related responses, and suppresses artifact-induced activation prior to feature reconstruction. The main results are reported as the mean and standard deviation of the sample over three runs. Swin-WaveNet achieved a Dice score of 89.69% ± 0.51% and a 95th-percentile Hausdorff distance (HD95) of [Formula: see text] pixels on the International Skin Imaging Collaboration (ISIC) 2018 dataset. Without fine-tuning on ISIC 2017, it obtained a Dice score of 91.69% ± 1.36% and an HD95 of [Formula: see text] pixels. On Kvasir-SEG, the model achieved a Dice score of 91.63% ± 0.57% and an HD95 of [Formula: see text] pixels. In a single-run component ablation, the complete model improved the Dice score by 0.96 percentage points and reduced HD95 by 1.16 pixels relative to the internal baseline. These results indicate that frequency-selective feature processing can help preserve weak or interrupted lesion boundaries.
Related
- UniMedSeg: Unified Multi-Modal Medical Image Segmentation via Uncertainty Guided Encoding and Frequency-Spatial Dual-Prompt Decoding
- UGB-SegNet: Uncertainty-Guided Multi-Scale Attention and Boundary-Aware Learning for Breast Ultrasound Lesion Segmentation
- Evaluation of boundary-aware UNet++ for low-grade glioma segmentation
- Boundary-aware multi-scale residual learning with hybrid radiomic–deep fusion for lung tumor segmentation and classification
- AMF-U-Net: an adaptive multimodal fusion residual attention 3D U-Net for boundary-aware brain tumour segmentation
BibTeX
@article{dan2026swinwavenet,
title = {{Swin-WaveNet: frequency-guided boundary restoration for robust skin lesion segmentation.}},
author = {Zhiping Dan and Run Zhang and Tinglong Tang and Yirong Wu and Xiaosheng Yu and Shuifa Sun and Ben Wang},
journal = {Visual computing for industry, biomedicine, and art},
year = {2026},
month = oct,
doi = {10.1186/s42492-026-00235-z},
url = {https://europepmc.org/article/MED/42895593},
note = {CC BY},
}
RIS
TY - JOUR
TI - Swin-WaveNet: frequency-guided boundary restoration for robust skin lesion segmentation.
AU - Zhiping Dan
AU - Run Zhang
AU - Tinglong Tang
AU - Yirong Wu
AU - Xiaosheng Yu
AU - Shuifa Sun
AU - Ben Wang
PY - 2026
DA - 2026/10/19
JO - Visual computing for industry, biomedicine, and art
DO - 10.1186/s42492-026-00235-z
UR - https://europepmc.org/article/MED/42895593
AB - Accurate skin lesion segmentation remains difficult when dermoscopic boundaries are fragmented by hair occlusion, uneven illumination, or weak local contrast. To address this problem, the authors propose Swin-WaveNet, a frequency-guided encoder-decoder that combines a Swin Transformer backbone with wavelet decomposition and selective state-space modeling. Rather than passing skip features directly to the decoder, the proposed method separates low-frequency semantic information from high-frequency details, selectively refines boundary-related responses, and suppresses artifact-induced activation prior to feature reconstruction. The main results are reported as the mean and standard deviation of the sample over three runs. Swin-WaveNet achieved a Dice score of 89.69% ± 0.51% and a 95th-percentile Hausdorff distance (HD95) of [Formula: see text] pixels on the International Skin Imaging Collaboration (ISIC) 2018 dataset. Without fine-tuning on ISIC 2017, it obtained a Dice score of 91.69% ± 1.36% and an HD95 of [Formula: see text] pixels. On Kvasir-SEG, the model achieved a Dice score of 91.63% ± 0.57% and an HD95 of [Formula: see text] pixels. In a single-run component ablation, the complete model improved the Dice score by 0.96 percentage points and reduced HD95 by 1.16 pixels relative to the internal baseline. These results indicate that frequency-selective feature processing can help preserve weak or interrupted lesion boundaries.
SN - 2096-496X
SN - 2524-4442
ER -