Hybrid Deep Features and Texture Features for Melanoma Classification

Authors

  • R.D Seeja , Dr. A. Suresh

Abstract

Melanoma is one of the dangerous kinds of skin cancer. Automatic melanoma classification system is very essential to lower the mortality rate by identifying the disease at earlier stages and accurate diagnosis. Even though there are many computerised methods for skin lesion classification, deep learning have been revealed to be greater over conventional methods. The main aim of this research work is to improve the melanoma classification performance using hybrid deep and texture features. Initially, the skin lesion area is segmented using a Convolutional Neural Network (CNN) based U-net algorithm. Features are extracted from segmented lesion by using VGG-16 and Local Binary Pattern. Also, these two features are combined to get more accurate classification. Finally, classification is done by using support vector machine classifier. Experiments are evaluated on ISBI-2016 and ISBI-2019 datasets. Experimental results show that, the hybrid features achieved 86.53% of accuracy in ISBI-2016 dataset and 93.5% of accuracy in ISBI-2019 dataset. In addition, the proposed approach has increase the performance of melanoma detection compared to previous studies.

Published

2020-12-01

Issue

Section

Articles