Skin Lesion Classification Based on Convolutional Neural Network

Skin cancer is one of the most common cancers, and its early detection can have a huge impact on outcomes. Deep learning, especially convolutional neural networks, perform well in processing massive amounts data, image data classifying skin cancer. In this paper, networks are mainly used to diagnose classify 7 types lesions, including melanoma, basal cell carcinoma, melanocytic nevus, actinic keratosis, intraepithelial benign keratinoid dermatofibroma, vascular lesions. First, characteristics lesion images analyzed, using technology sampling preprocess images. Then training parameters imageNet network adjusted through idea transfer learning InceptionV3, ResNet50, DenseNet201, other classification. Furthermore, different models built for order validate classification performance various models, paper adopts ISIC 2017 HAM10000 dataset experiments. The experimental results show that proper preprocessing necessary when applying CNN 224*224 images, classical deep with DenseNet201 achieved remarkable accuracy 99.12% 86.91% testing.