A Transfer-Learning-Based Novel Convolution Neural Network for Melanoma Classification

Skin cancer is one of the most common human malignancies, which generally diagnosed by screening and dermoscopic analysis followed histopathological assessment biopsy. Deep-learning-based methods have been proposed for skin lesion classification in last few years. The major drawback all that they require a considerable amount training data, poses challenge classifying medical images as limited datasets are available. problem can be tackled through transfer learning, model pre-trained on huge dataset utilized fine-tuned per domain. This paper proposes new Convolution neural network architecture to classify lesions into two classes: benign malignant. Google Xception used base top layers added then fine-tuned. optimized using various optimizers achieve maximum possible performance gain classifier output. results ISIC archive data achieved highest accuracy 99.78% Adam LazyAdam optimizers, validation test 97.94% 96.8% RMSProp, HAM10000 utilizing RMSProp optimizer, prediction 98.81% 91.54% respectively, when compared other models.