IESBU-Net: A Lightweight Skin Lesion Segmentation UNet with Inner-Module Extension and Skip-Connection Bridge

Skin disease is one of the most common human diseases. In order to improve the segmentation of automatic skin lesions, some pioneering work often uses more complex modules to improve the segmentation performance. However, the models with high computational complexity are difficult to be applied to the realistic medical scenarios with limited computational resources. To solve this problem, we propose a light-weight model, IESBU-Net. It achieves a balance between cost of parameters and computational complexity in the segmentation of skin lesions. Briefly, we propose three modules: (1) the introduction of depthwise separable convolutions by LSR in shallow encoders reduces computational complexity and expands the receptive field for protecting primary feature information. (2) CA collects dense global feature information in the context-rich encoder bottleneck layer. (3) SCB adds a bridge in the process of feature fusion between encoder and decoder, which is used to smooth the semantic gap between encoder and decoder caused by skip-connection. We evaluated the proposed model on the ISIC 2017 and ISIC 2018 datasets. The experimental results show that the model achieves a good balance between the number of parameters, computational complexity and performance, and improves the performance of skin lesion segmentation significantly.