EGFNet: Efficient guided feature fusion network for skin cancer lesion segmentation

Abstract

Melanoma is the leading cause of death from skin cancer, and the number is increasing every year. However, automated segmentation of melanoma remains a challenging problem due to the great variation in shape, colour and texture of melanoma. Moreover, with the development of mobile devices, achieving higher performance segmentation on embedded devices deserves further research. To address the above issues, this paper proposes a lightweight network for skin lesion segmentation with guided learning based on the attention mechanism, which not only ensures image segmentation accuracy using an efficient feature fusion module, but also effectively reduces the complexity of the model. Extensive experiments on the ISIC2017 dataset validate that EGFNet achieves very competitive results in terms of objective metrics.