Structure-preserving lightweight skin lesion segmentation with dual-resolution attention and boundary-aware fusion
Abstract
Objective Dermoscopic skin lesion segmentation suffers from low contrast, artifacts, and blurred boundaries in lightweight models due to lossy downsampling and rigid tokenization. We propose Structure-Preserving Representation Learning (SP-RL), a lightweight framework designed to preserve geometric topology and continuity for stronger representation learning. Methods SP-RL integrates four components: 1) Structure-Preserving Sampling Strategy (SPSS) using a bijective pixel-folding mechanism to redistribute spatial details into channel dimensions; 2) Lite Multi-Scale Aggregation Block (LMA-Block) to expand receptive fields at high resolutions; 3) Dual-Resolution Attention (DR-Atn) maintaining high-resolution queries while compressing keys/values via hierarchical pooling; and 4) Dual-Gated Multi-Scale Fusion (DG-MSF) to suppress low-level artifacts and inject noise-resilient Sobel gradient priors into progressive deconvolution paths. Results Evaluations across 23 models show SP-RL achieves a 0.8976 Dice score and the lowest HD95 (15.1829) on ISIC2018 among lightweight alternatives. It exhibits superior robustness across four public datasets (ISIC 2016/2017/2018, PH 2 ) and unseen domains like HAM10000 under severe noise and focus blur. Conclusion SP-RL offers an optimal trade-off between boundary accuracy, reliability, and efficiency. By retaining high-frequency energy and domain-invariant morphology, it provides a dependable solution for resource-constrained diagnosis.
Related
- Automated corneal endothelial cell detection and segmentation in specular microscopy images using Attention and Faster U-Net
- AMF-U-Net: an adaptive multimodal fusion residual attention 3D U-Net for boundary-aware brain tumour segmentation
- Hybrid feature pyramid attention and capsule network (FpaNet+CapsNet) for enhanced spatial and contextual feature extraction for improved brain tumor segmentation
- TiLCA-Net: nuclei segmentation and classification in pathological images via task-specialized attention and efficient decoding
- Efficient and real-time polyp segmentation: An empirical evaluation of lightweight Mamba integration strategies
BibTeX
@article{yang2026structurepreserving,
title = {{Structure-preserving lightweight skin lesion segmentation with dual-resolution attention and boundary-aware fusion}},
author = {Xu Yang and Duwei Dai and Caixia Dong and Liang Wang and Pengyu Ren},
journal = {Digital Health},
year = {2026},
month = oct,
doi = {10.1177/20552076261502184},
url = {https://doi.org/10.1177/20552076261502184},
note = {CC BY},
}
RIS
TY - JOUR
TI - Structure-preserving lightweight skin lesion segmentation with dual-resolution attention and boundary-aware fusion
AU - Xu Yang
AU - Duwei Dai
AU - Caixia Dong
AU - Liang Wang
AU - Pengyu Ren
PY - 2026
DA - 2026/10/29
JO - Digital Health
DO - 10.1177/20552076261502184
UR - https://doi.org/10.1177/20552076261502184
AB - Objective Dermoscopic skin lesion segmentation suffers from low contrast, artifacts, and blurred boundaries in lightweight models due to lossy downsampling and rigid tokenization. We propose Structure-Preserving Representation Learning (SP-RL), a lightweight framework designed to preserve geometric topology and continuity for stronger representation learning. Methods SP-RL integrates four components: 1) Structure-Preserving Sampling Strategy (SPSS) using a bijective pixel-folding mechanism to redistribute spatial details into channel dimensions; 2) Lite Multi-Scale Aggregation Block (LMA-Block) to expand receptive fields at high resolutions; 3) Dual-Resolution Attention (DR-Atn) maintaining high-resolution queries while compressing keys/values via hierarchical pooling; and 4) Dual-Gated Multi-Scale Fusion (DG-MSF) to suppress low-level artifacts and inject noise-resilient Sobel gradient priors into progressive deconvolution paths. Results Evaluations across 23 models show SP-RL achieves a 0.8976 Dice score and the lowest HD95 (15.1829) on ISIC2018 among lightweight alternatives. It exhibits superior robustness across four public datasets (ISIC 2016/2017/2018, PH 2 ) and unseen domains like HAM10000 under severe noise and focus blur. Conclusion SP-RL offers an optimal trade-off between boundary accuracy, reliability, and efficiency. By retaining high-frequency energy and domain-invariant morphology, it provides a dependable solution for resource-constrained diagnosis.
SN - 2055-2076
ER -