Abstract

Info

Title:: Skin lesion segmentation method combining channel weight update and dense residual pyramid spatial attention

Author(s):: CHEN Jingjing1; LI Xiaoxia1; 2; L?Nianzu1; 1. School of Information Engineering, Southwest University of Science and Technology, Mianyang 621000, China 2. Sichuan Key Laboratory of Robotics in Special Environment, Mianyang 621000, China

Keywords:: skin lesion segmentation U-Net attention mechanism boundary loss function

Abstract:: The segmentation of skin lesions is critical for the computer-aided diagnosis of melanoma. A novel skin lesion segmentation method is proposed based on U-Net for extracting skin lesions more accurately. The proposed method adopts the channel weight update module and the dense residual pyramid spatial attention module to extract effective information from channels and space, highlight lesion features and suppress irrelevant features, thereby improving the accuracy of the network for the segmentation of pathological regions. In addition, a weighted boundary loss function is constructed to reduce the loss of lesion edge features through strong supervision on lesion contours. The experiment results show that the proposed method achieves Dice coefficients of 91.3% and 92.2% on ISIC 2018 and PH2 dermoscopic image data sets, respectively, which are improved by 5.0% and 4.3% as compared with U-Net.