Image Segmentation of Skin Lesions Based on Attention Mechanism and Deep Convolutional Neural Network

With the increasing fatality rate of melanoma, accurate segmentation of lesion area in skin disease image plays an important role in the diagnosis of skin cancer. Computer-aided diagnosis system can provide objective opinions for the diagnosis and treatment of melanoma. The existing skin disease images are mostly obtained by professional dermoscope instruments, and the judgment of the skin lesion area is mainly from the judgment of professional doctors, but this method will be affected in both efficiency and accuracy. In the feature analysis of image data sets, there are some problems, such as a small number of image labeled samples, large diversity of skin lesion area and no obvious difference between lesion area and background. In order to improve the ability of feature extraction, a neural network with attention mechanism is proposed. The algorithm is based on Mask R-CNN network and channel attention mechanism to improve the importance of acquiring each feature channel and enhance the useful features. Experimental results on data sets containing different degrees of melanoma show that the algorithm can achieve better performance than the traditional Mask R-CNN.