Optimized Convolutional Neural Network Models for Skin Lesion Classification

Skin cancer is one of the most severe diseases, and medical imaging among main tools for diagnosis. The images provide information on evolutionary stage, size, location tumor lesions. This paper focuses classification skin lesion considering a framework four experiments to analyze performance Convolutional Neural Networks (CNNs) in distinguishing different CNNs are based transfer learning, taking advantage ImageNet weights. Accordingly, each experiment, workflow stages tested, including data augmentation fine-tuning optimization. Three CNN models DenseNet-201, Inception-ResNet-V2, Inception-V3 proposed compared using HAM10000 dataset. results obtained by three demonstrate accuracies 98%, 97%, 96%, respectively. Finally, best model tested ISIC 2019 dataset showing an accuracy 93%. methodology represents helpful tool accurately diagnose disease.