Skin Lesion Classification Using Deep Convolutional Neural Network and Transfer Learning Approach

Abstract

For biomedical image processing and prediction of skin diseases, deep learning methods are playing a very significant role in better decision making. This paper has proposed an automatic classification system of images containing a skin lesion as malignant or benign. In this method the transfer learning and a pre-trained deep learning network are implemented. In this proposed work transfer learning is applied to VGGNet architecture by replacing the last layer by a softmax layer for the classification of two different lesions (malignant and benign). Fine-tuning, data augmentations, and cross-validations are also added to the method. After evaluating the performances of the proposed method on the testing set of the ISIC dataset the method has achieved a significantly higher classification accuracy rate of 98.02%, the sensitivity of 98.10%, and Specificity of 97.05%.