Transfer Learning Based Method for Two-Step Skin Cancer Images Classification

Abstract

Classification of skin cancer images is an important research challenge because the number of people is expected to increase significantly in the next years and consequently the number of people that might have skin cancer in one phase of life will increase accordingly. Some types of skin cancer can be treated successfully if they are detected in the early stages and thus the study of the skin cancer images using the latest technological innovations might lead to better results than in the case when traditional methods are applied. In this article is presented a method for the classification of skin cancer images that consists of two steps and which is based on transfer learning and deep learning. The classification models are developed in Python using the PyTorch machine learning library and the dataset used as experimental support for testing and validating the transfer learning based method is Human Against Machine with 10000 training images (HAM10000) dataset. In the first step the accuracy of the prediction model for testing data is 85% and in the second step the accuracy of the prediction model for testing data is 75%.