LCAUnet: A skin lesion segmentation network with enhanced edge and body fusion

Qisen Ma,Keming Mao,Gao Wang,Lisheng Xu,Yuhai Zhao

2023-05-01

Abstract:Accurate segmentation of skin lesions in dermatoscopic images is crucial for the early diagnosis of skin cancer and improving the survival rate of patients. However, it is still a challenging task due to the irregularity of lesion areas, the fuzziness of boundaries, and other complex interference factors. In this paper, a novel LCAUnet is proposed to improve the ability of complementary representation with fusion of edge and body features, which are often paid little attentions in traditional methods. First, two separate branches are set for edge and body segmentation with CNNs and Transformer based architecture respectively. Then, LCAF module is utilized to fuse feature maps of edge and body of the same level by local cross-attention operation in encoder stage. Furthermore, PGMF module is embedded for feature integration with prior guided multi-scale adaption. Comprehensive experiments on public available dataset ISIC 2017, ISIC 2018, and PH2 demonstrate that LCAUnet outperforms most state-of-the-art methods. The ablation studies also verify the effectiveness of the proposed fusion techniques.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The paper primarily focuses on addressing the problem of skin lesion segmentation, particularly the accurate segmentation of skin lesions in dermoscopic images, which is crucial for the early diagnosis of skin cancer and improving patient survival rates. However, due to the irregularity of lesion areas, blurred boundaries, and other complex interfering factors, automatically distinguishing lesions from healthy skin remains a challenge.

To tackle these challenges, the authors propose a new network architecture—LCAUnet. The main innovation of LCAUnet lies in its enhanced edge and main body fusion capability, which is often overlooked in traditional methods. Specifically, LCAUnet includes the following key components:

1. **Dual-branch Encoder**: One branch is used for edge segmentation (based on Convolutional Neural Networks, CNNs), and the other for main body segmentation (based on Transformer architecture). This design allows the network to simultaneously extract edge information and main body features.

2. **Local Cross Attention Fusion Module (LCAF)**: This module uses local cross attention operations during the encoding phase to fuse edge and main body feature maps at the same level. This helps to more accurately integrate complementary features and reduce computational complexity.

3. **Prior-guided Multi-scale Fusion Module (PGMF)**: This module employs prior knowledge for multi-scale adaptive feature integration to better fuse features at different scales.

Through these techniques, LCAUnet outperforms most existing advanced methods on public datasets ISIC 2017, ISIC 2018, and PH2. Additionally, ablation experiments also validate the effectiveness of the proposed fusion techniques.

In summary, LCAUnet aims to improve the accuracy of skin lesion segmentation by enhancing the fusion of edge and main body features, particularly excelling in handling irregular shapes and difficult-to-detect boundaries.