EM-Net: Effective and Morphology-aware Network for Skin Lesion Segmentation

Expert Systems with Applications, 2025
Kaiwen Zhu1,2,3*, Yuezhe Yang3,4*, Yonglin Chen1,3*†, Ruixi Feng5, Dongping Chen4, Bingzhi Fan4, Nan Liu4, Ying Li6, Xuewen Wang6
*Equal Contribution †Corresponding author
1Anhui Jianzhu University, 2Southeast University, 3Anhui Provincial International Joint Research Center for Advanced Technology in Medical Imaging, 4Anhui University, 5Stony Brook Institute at Anhui University, 6Zhejiang University School of Medicine
Paper DOI Code

Abstract

Dermoscopic images are essential for diagnosing various skin diseases, as they enable physicians to observe subepidermal structures, dermal papillae, and deeper tissues otherwise invisible to the naked eye. However, segmenting lesions in these images is challenging due to their irregular boundaries and significant variability in lesion characteristics.

We propose an effective and morphology-aware network that utilizes a hybrid feature extractor combining CNN and ViT architectures. EM-Net introduces a boundary delineation component based on non-convex optimization to learn general representations and accurately delineate lesion boundaries, enhancing detail extraction.

The model also integrates a few-shot domain generalization module to improve adaptability across datasets. Validation on publicly available dermoscopic image datasets, including ISIC, PH², PAD-UFES-20, and the University of Waterloo skin cancer database, demonstrates state-of-the-art performance in Dice, Acc, Pre, IoU, and Re.

Highlights

  • Propose a Morphology-aware Module to capture lesion contours, corners, and boundary details.
  • Integrate few-shot domain generalization to improve cross-dataset robustness.
  • Combine CNN and Vision Transformer features in a hybrid encoder-decoder segmentation framework.
  • Provide additional segmentation masks for PAD-UFES-20 cellphone dermatology images.

Method

EM-Net framework

The EM-Net framework includes the Morphology-aware Module, Global Attention Module, Bridge Fusion Module, and the main segmentation network.

Morphology-aware Module

Morphology-aware module illustration

The Morphology-aware Module captures structural details in dermoscopic images and helps EM-Net better preserve lesion boundaries during segmentation.

Results

ISIC segmentation comparison

ISIC Dataset

ISIC 2016 and PH2 segmentation comparison

ISIC 2016 + PH²

Mobile phone dataset segmentation comparison

Mobile Phone Dataset

Citation

@article{zhu2025emnet,
  title={EM-Net: Effective and morphology-aware network for skin lesion segmentation},
  author={Zhu, Kaiwen and Yang, Yuezhe and Chen, Yonglin and Feng, Ruixi and Chen, Dongping and Fan, Bingzhi and Liu, Nan and Li, Ying and Wang, Xuewen},
  journal={Expert Systems with Applications},
  volume={285},
  pages={127668},
  year={2025},
  doi={10.1016/j.eswa.2025.127668},
  publisher={Elsevier}
}