Abstract:Skin melanoma is one of the most common malignant tumors originating from melanocytes, and the incidence of the Chinese population is showing a continuous increasing trend. Early and accurate diagnosis of melanoma has great significance for guiding clinical treatment. However, the symptoms of malignant melanoma are not obvious in the early stage. It is difficult to be diagnosed with human observation. Meanwhile, it is easy to spread due to missed diagnosis. In order to accurately diagnose melanoma, end-to-end skin lesion attribute segmentation framework is presented in this paper. It is applied to facilitate the digitalization process of attributes segmentation. The framework was improved on the U-Net construction that use the channel context feature fusion module between the encoder and decoder to further merge context information. A dual-domain attention module is proposed to get more effective information from the feature map. It shows that the proposed method effectively segments the lesion attributes and achieves good result in the ISIC2018 task2 dataset.

Channel Context and Dual-Domain Attention Based U-Net for Skin Lesion Attributes Segmentation