Automatic Detection of Melanoma Skin Cancer from Dermoscopy Images based on Features Fusion

Abstract

Skin cancer is a general health problem. It can occur in young and adults. The most deadly and prevalent kind is melanoma. It occurs in melanocyte cells, which create melanin, and spreads to other body areas. It is necessary to discover melanoma at an early stage to decrease the mortality rate. Traditional clinical methods required dermatologists to check all patients, this way consume time, cost, and effort. Automated detection helps to obtain accurate results. Extraction and selection of features for melanoma detection from dermoscopy images is a challenging task. This research aims to extract efficient features by fusion-handcrafted features Gray Level Co-occurrence matrix, local binary patterns, and pre-trained convolution neural network features (GoogleNet, AlexNet, ResNet18, and DensNet201). Experiments have been conducted to explore if the fusion of two powerful features together will lead to enhanced performance or not. The proposed system used PH2 and ISIC2017 datasets. According to the experimental results, combining deep with handcrafted features enhances the performance of classification. when compared to using just deep or handcrafted features alone.