Cutting Effect on Classification Using Nasnet Architecture

Abstract

Malignant melanoma is the most dangerous and lethal skin cancer type. In time and early diagnosis increases the possibility of successful treatment. Studies done in recent years show that skin cancer diagnosis can be done by using computer aided diagnosis systems. In this study, a classification using a dataset with two classes which are malign and benign melanom is realized. Nasnet deep learning architecture is used for the classification. Two different experiments are done in this study. While first experiment classifies dataset directly by using Nasnet architecture, second experiment does classification by first creating 8 images from train and validation part of dataset and starting training by using this new dataset. While an accuracy rate of 82.94% is get without cutting operation, an accuracy rate of 86.49% is get with cutting operation. A better classification performance is reached with cutting operation.