Research on Dermoscopic Segmentation based on Multi-scale Convolutional Neural Network

摘要: Abstract The skin is an important organ of the human body, and major diseases represented by melanoma are difficult to find treat, which directly threaten patient’s life safety. Therefore, rapid accurate dermatoscopy diagnosis[1] particularly important. lack experienced dermatologists limits large-scale early diagnosis or screening melanoma. Based on above problems, this paper uses method multi-scale convolutional neural network segment Dermoscopic image means segmentation in artificial intelligence help medical personnel achieve reliable diagnostic reference. In overall network, first enhances contrast original through preprocessing, data enhancement increase dataset. model training phase, adopts basic framework U-Net[2] network. Different from U-Net networks, introduces features fusion mechanisms deeply mine fuse dermoscopic features. Secondly, Focal Loss as fundus loss function, finds optimal parameters function a random grid search algorithm. paper, for problem positive negative sample imbalance images, proposes adaptive Channel-Wise Attention mechanism adaptively weight feature maps.