A Hybrid Model for Skin Disease Classification using Transfer Learning

Abstract

Worldwide, around two people die each hour due to skin cancer. The disease is normally originated by expose to sunrays. Early detection is very important to prevent it from spreading. The traditional method of detecting skin cancer is through a procedure known as Biopsy. This is an invasive and time-consuming procedure that involves removing the skin cells. With the advancement of imaging techniques, early detection of skin cancer can be made possible. A study has been conducted to develop two deep learning architectures that can automatically detect skin cancer using 3700 clinical images. One of the architectures is based on the AlexNet framework, which is a transfer learning algorithm. The other one uses a hybrid structure that combines the long short term memory and the temporal properties of the images. The first architecture, which is based on the AlexNet framework, has an accuracy of 99.25%. However, the second hybrid structure, which is a combination of the long-term memory and the temporal properties, has an accuracy of 99.75%. The results of the study contribute to the field of the deep structural model.