Effective Attention-Guided Multi-Scale Medical Network for Skin Lesion Segmentation

TL;DR

The paper tackles accurate skin lesion segmentation under irregular shapes and low-contrast conditions. It introduces EAM-Net, a lightweight encoder–decoder framework that combines Cross-Mix Attention Module (CMAM), Multi-Resolution Multi-Channel Fusion (MRCF), External Attention Bridge (EAB), and multi-scale encoding/decoding fusion to capture cross-scale context and preserve decoder information. Extensive experiments on ISIC2018 and PH2 demonstrate state-of-the-art performance with a compact model, highlighting improved boundary accuracy and robustness. The work suggests strong potential for clinical deployment and outlines avenues for broader generalization to other medical imaging tasks.

Abstract

In the field of healthcare, precise skin lesion segmentation is crucial for the early detection and accurate diagnosis of skin diseases. Despite significant advances in deep learning for image processing, existing methods have yet to effectively address the challenges of irregular lesion shapes and low contrast. To address these issues, this paper proposes an innovative encoder-decoder network architecture based on multi-scale residual structures, capable of extracting rich feature information from different receptive fields to effectively identify lesion areas. By introducing a Multi-Resolution Multi-Channel Fusion (MRCF) module, our method captures cross-scale features, enhancing the clarity and accuracy of the extracted information. Furthermore, we propose a Cross-Mix Attention Module (CMAM), which redefines the attention scope and dynamically calculates weights across multiple contexts, thus improving the flexibility and depth of feature capture and enabling deeper exploration of subtle features. To overcome the information loss caused by skip connections in traditional U-Net, an External Attention Bridge (EAB) is introduced, facilitating the effective utilization of information in the decoder and compensating for the loss during upsampling. Extensive experimental evaluations on several skin lesion segmentation datasets demonstrate that the proposed model significantly outperforms existing transformer and convolutional neural network-based models, showcasing exceptional segmentation accuracy and robustness.

Figures (8)

• Figure 1: Some typical challenging skin lesions in the ISIC2018 and PH2 datasets. The first two are irregular lesions, the third is small lesions, the fourth and fifth are lesions with low contrast to the background, and the last is lesions with hair blocked.
• Figure 2: The main architecture of EAM-Net.The comments at the bottom represent the actual operations of the modules not labeled in the figure.
• Figure 3: Main architecture of the MRCF module.
• Figure 4: Implementation of the Split module in the MRCF module.
• Figure 5: The architecture of the EAB module.